Building Information Modeling and AI Algorithms for Optimizing Energy Performance in Hot Climates: A Comparative Study of Riyadh and Dubai

Abstract

In response to increasing global temperatures and energy demands, optimizing buildings’ energy efficiency, particularly in hot climates, is an urgent challenge. While current research often relies on conventional energy estimation methods, there has been a decrease in the efforts dedicated to leveraging AI-based methodologies as technology advances. This implies a dearth of multiparameter examinations in AI-driven extreme case studies. For this reason, this study aimed to enhance the energy performance of residential buildings in the hot climates of Dubai and Riyadh by integrating Building Information Modeling (BIM) and Machine Learning (ML). Detailed BIM models of a typical residential villa in these regions were created using Revit, incorporating conventional, modern, and green building envelopes (BEs). These models served as the basis for energy simulations conducted with Green Building Studio (GBS) and Insight, focusing on crucial building features such as floor area, external and internal walls, windows, flooring, roofing, building orientation, infiltration, daylighting, and more. To predict Energy Use Intensity (EUI), four ML algorithms, namely, Gradient Boosting Machine (GBM), Random Forest (RF), Support Vector Machine (SVM), and Lasso Regression (LR), were employed. GBM consistently outperformed the others, demonstrating superior prediction accuracy with an R² of 0.989. This indicates that the model explains 99% of the variance in EUI, highlighting its effectiveness in capturing the relationships between building features and energy consumption. Feature importance analysis (FIA) revealed that roofs (29% in Dubai scenarios (DS) and 40% in Riyadh scenarios (RS)), external walls (19% in DS and 29% in RS), and windows (15% in DS and 9% in RS) have the most impact on energy consumption. Additionally, the study explored the potential for energy optimization, such as cavity green walls and green roofs in RS and double brick walls with VIP insulation and green roofs in DS. The findings of the paper should be interpreted in light of certain limitations but they underscore the effectiveness of combining BIM and ML for sustainable building design, offering actionable insights for enhancing energy efficiency in hot climates.

1. Introduction

Optimizing energy performance and efficiency has been a long-standing concern for policymakers, real estate developers, and designers worldwide [1]. It is also a crucial aspect of the global sustainability agenda, reflecting a robust global focus [2]. For instance, energy efficiency is a critical element of the European Union’s sustainability agenda and is expected to remain a significant focus until 2050 [3,4]. According to the European Environment Agency, improving energy efficiency is a practical short-term approach to reducing greenhouse gas emissions, enhancing energy security, and promoting sustainable development [4]. Similarly, in Australia, residential buildings are accountable for approximately a quarter of the nation’s total electricity consumption and more than 10% of its total carbon emissions [5,6]. On the global scale, energy consumption contributes up to 40% [7]. The study of energy-related carbon emissions highlights the urgent need for increased energy efficiency [8].

Notably, the growing global awareness of environmental sustainability has brought significant attention to the energy efficiency of buildings, especially in the Gulf regions with hot climates, including Saudi Arabia and the United Arab Emirates. For example, the Kingdom of Saudi Arabia uses almost 35% of its daily oil output for electrical power use, with an annual rise of roughly 5% to 8% [9,10]. Moreover, buildings consume over 80% of energy annually [11]. Notably, residential buildings alone consume 50% of the total energy the building stock uses [12]. In Saudi Arabia, recent statistics by the General Authority for Statistics in 2024 revealed that the building numbers reached up to 12 million building units. The buildings in Riyadh constitute 25% of all buildings, and residential buildings represent 72.7% of all buildings in Riyadh [13]. This percentage has likely increased due to the city’s ongoing development and population growth. This, in turn, necessitated developing the Saudi Energy Conservation Code-Low Rise (SBC 602) to rationalize residential building energy consumption, supporting the Kingdom’s overarching objectives of reducing carbon emissions, lowering energy consumption, and advancing sustainable development [14]. Notably, the economic and climatic conditions of both cities, Riyadh and Dubai, are comparable, as are the governments’ efforts to reduce the energy consumption of residential buildings. This prompted us to focus exclusively on the two promising cities in the hot climate region.

Despite the ongoing efforts of global and local institutions to reduce energy consumption and standardize energy performance, policymakers, designers, developers, and governments continue encountering substantial challenges in achieving energy efficiency. This is because of the underutilization of new technologies, financial constraints, and difficulties implementing consistent regulations across various sectors and regions [4].

While there has been progress in achieving the United Nations Sustainable Development Goals (SDGs), several EU member states are projected to fall short of the 2030 targets, indicating immediate corrective actions are needed [15]. Recognizing these challenges, this study supports SDG 7: Affordable and Clean Energy, SDG 11: Sustainable Cities and Communities, and SDG 13: Climate Action. This study aims to reduce energy consumption and enhance sustainability by optimizing energy performance in residential buildings.

In light of these ongoing challenges and the need for immediate action, new technologies enable designers to tackle persistent issues using novel and innovative materials, thereby generating creative solutions that yield better outcomes. Several studies have used technologies such as computer-based simulation programs, i.e., DesignBuilder and Energyplus. However, few studies have integrated ML with BIM for building energy performance.

The structure of this paper is organized as follows: Section 2, Literature Review, highlights current research in three main areas: BEs and insulation materials, green buildings, and advanced simulation and optimization techniques. Section 3, Methodology, describes the methodology used in the study, including model development, data collection, and the various methods employed to evaluate building energy performance and optimization; Section 4, Results, presents a detailed analysis of the performance of ML algorithms in predicting EUI, and it evaluates the most important building features and optimization strategies; Section 5, Discussion, discusses the implications of the findings, contributions to the field of building energy performance, and limitations and potential areas for future research; and Section 6, Conclusion, summarizes the key findings.

2. Literature Review

2.1. Residential Building Envelopes (BEs) and Insulation Materials in Hot Climates

BEs and insulation materials are pivotal in enhancing energy efficiency in hot climates. Concerning the residential buildings in Saudi Arabia, Al-Qahtani and Elgizawi [16] conducted a case study research in Riyadh, evaluating the energy-saving potential of passive cooling strategies such as shading devices, upgraded glazing, wall and roof insulation, and green roofs using DesignBuilder software. The results revealed that adding insulation to the outer walls and roofs significantly reduced cooling loads by approximately 57.6%. Again, using DesignBuilder, this study aligns with the findings of Alyami [17], which demonstrated that incorporating thermal insulation in both walls and roofs decreased energy use by about 45% overall. However, Al-Qahtani and Elgizawi [16] found that shading devices and upgraded glazing contributed to energy savings but to a lesser extent. Green roofs provided additional thermal benefits, though their effectiveness was somewhat diminished when combined with other strategies. Similarly, Aldersoni et al. [18] utilized DesignBuilder software to analyze the impact of four passive strategies—outdoor green areas, thermal mass walls, Window–Wall Ratio (WWR), and shading devices—on the energy performance of traditional houses in Riyadh, Saudi Arabia. Results indicated that thermal mass walls and outdoor green areas were particularly effective in reducing cooling energy demand by up to 25%. Adjusting the WWR and incorporating shading devices also contributed to energy savings by minimizing direct solar gain and improving natural ventilation.

Alsaqabi et al. [19] evaluated various insulation materials in four residential projects at Qassim University, Saudi Arabia. They used Integrated Environmental Solutions—Virtual Environment (IES-VE) and SimaPro, IMPACR2002+, and found that rock wool (RW) and glass wool (GW) significantly reduce energy consumption due to their superior thermal properties. Multi-case study research conducted by Alyami et al. [20] investigated the effects of five insulation materials: polyurethane board (PU), expanded polystyrene (EPS), GW, urea-formaldehyde foam, and expanded perlite in five cities, i.e., Abha, Dammam, Riyadh, Najran, and Tabuk using IES-VE. Such a study highlights that PU provided the highest energy savings with potential reductions in energy demand by up to 14%.

In Bahrain, Dhaif and Stephan [21] compared Structural Insulated Panels (SIPs) with hollow concrete blocks. The findings showed that SIPs, with an insulating foam core between oriented strand board (OSB) skins, provide superior thermal insulation and reduce cooling energy use by 20.6% due to better thermal resistance and air tightness.

A study in New Aswan City, Egypt, investigated nano vacuum insulation panels (VIPs) and nano gel glazing, achieving a 47.6% total energy reduction compared to conventional materials [22]. In eight hot desert climates, Sovetova et al. [23] assessed 13 different Phase Change Materials (PCMs) in different cities like Mecca, Dubai, Abu-Dhabi Faisalabad, Jodhpur, Nouakchott, Cairo, and Biskra. They utilized EnergyPlus simulations showing energy consumption reductions of up to 34.26%.

2.2. Green Buildings in Hot Climates

Green building practices are increasingly being recognized for their potential to reduce energy consumption and environmental impact. Mahmoud and Ismaeel [24] developed sustainable roof design guidelines for hot arid climates, finding that while green and sloped roofs contributed to energy savings, reflective roofs proved most effective in reducing solar heat gain and improving thermal comfort in hot arid climates. Ragab and Abdelrady [25] investigated how green roofs can reduce cooling energy demand in school buildings across Cairo, Alexandria, and Aswan. Using Design-Builder, they tested various green roof configurations with different soil depths and thermal conductivities. The study found that green roofs significantly decreased energy consumption for cooling, with the greatest savings in Aswan, by up to 39.74%. The most efficient green roof had a 0.1 m soil depth and 0.9 W/m-K thermal conductivity, effectively lowering air temperature and cooling energy demand. The study concludes that green roofs are highly effective in hot, arid climates for improving energy efficiency and thermal comfort. Andric, Kamal, and Al-Ghamdi [26] assessed the effectiveness of green roofs and green walls in mitigating climate change in Qatar’s extremely hot and dry climate. Using future climate projections and building modeling, they evaluated energy consumption in residential buildings for 2020, 2050, and 2080. The study found that traditional insulation methods, like EPS and energy-efficient windows, reduced energy consumption by up to 30%, while green roofs and walls achieved only around a 3% reduction. This limited impact is attributed to Qatar’s extreme temperatures, where the shading provided by green walls does not significantly reduce overall heat transfer through the building envelope (BE). The study concluded that while green infrastructure can enhance air quality and mitigate the urban heat island effect, its direct impact on energy savings is minimal compared to conventional insulation methods. Alyami et al. [17] investigated the feasibility of achieving Zero Energy Buildings (ZEB) in Saudi Arabia’s hot and dry climate by implementing various green building solutions. Using IES-VE software, the study simulated Building Energy Performance (BEP) and evaluated sustainable practices such as thermal mass, daylighting, natural ventilation, cavity walls, double-glazing, and solar panels. A building management system manages these strategies and shows significant potential in reducing energy consumption and achieving ZEB status. The study concluded that with appropriate design adaptations and renewable energy technologies, transforming residential buildings in Saudi Arabia into ZEB is feasible, supporting the country’s Vision 2030 sustainability goals. Pragati et al. [27] studied the effects of green roofs and walls on BEP in tropical climates using DesignBuilder software. They found up to a 10.5% reduction in total energy consumption and a 13% decrease in cooling demand in green buildings.

2.3. BIM, ML, and Optimization in Building Energy Efficiency

Amani and Soroush [28] evaluated the impact of BIM on optimizing energy consumption in residential buildings. By modeling various scenarios in Autodesk Revit and GBS, they reported significant energy cost savings of 58.23% for block D, 51.03% for block C, and 43.05% for the western lobby. Additionally, notable reductions in EUI were observed. The study highlighted that using BIM to optimize building parameters like orientation and material properties can reduce energy costs and emissions, although results may vary with different climates and building designs. Similarly, Gao et al. [29] reviewed the integration of BIM with Building Energy Modeling (BEM) to enhance energy efficiency in building designs. Their study discussed the limitations of conventional BEM, such as tedious model preparation and a lack of synchronization with the digital design process, and they emphasized the benefits of BIM-based BEM in improving efficiency and accuracy.

In addition to the advancements in BIM, the application of ML has also shown significant promise in enhancing building energy efficiency. Albatayneh [30] explored the optimization of BE parameters for a single-story residential building in Jordan, which has a semi-arid, warm Mediterranean climate. The study used DesignBuilder software for energy simulations and a genetic algorithm for optimization to reduce cooling and heating loads while maintaining thermal comfort. Key design variables included roof and floor insulation thickness, WWR, glazing type, and shading devices. Sensitivity analysis was conducted to assess their impact on energy performance. The optimization significantly reduced total energy consumption from 5225.97 kWh/year to 293.74 kWh/year, demonstrating the effectiveness of these strategies in enhancing residential energy efficiency. Furthermore, Chegari et al. [31] developed a multi-objective optimization method to improve indoor thermal comfort and energy performance in residential buildings in Morocco’s hot climate. The study combined Artificial Neural Networks (ANNs) with metaheuristic algorithms, including Multi-Objective Particle Swarm Optimization (MOPSO), Non-dominated Sorting Genetic Algorithm, and Multi-Objective Genetic Algorithm. Using TRNSYS software for dynamic thermal simulations, the method reduced thermal energy demands by up to 74.52% and improved indoor thermal comfort by 4.32%. MOPSO emerged as the most effective algorithm, optimizing BE design variables such as thermal transmission coefficients and shading devices.

Additionally, Egwim et al. [32] examined the performance of ML algorithms for predicting building energy efficiency in the UK, implementing a hybrid stacking ensemble approach that proved more effective than single algorithms. Likewise, Guo et al. [33] developed an ML-based method for designing energy-efficient forms for detached residential buildings, using the Grasshopper platform to integrate data sampling, performance simulation, and accuracy evaluation. They compared ANN and SVM algorithms, finding that ANN provided the highest accuracy for predicting energy consumption. Further supporting these findings, Seyedzadeh et al. [34] reviewed various ML techniques, noting the robustness of ANN and SVM in handling non-linear problems and enhancing prediction accuracy. Lastly, Barbaresi et al. [35] evaluated the performance of ML models for predicting building energy needs to reduce computational time in energy simulations. The study compared SVM, RF, and Extreme Gradient Boosting (XGB) models using a dataset of energy simulations for a winery building in Italy. The results showed that the XGB model provided the best accuracy and computational time performance, significantly reducing the need for extensive energy simulations.

2.4. Purpose and Significance

Many studies focus solely on energy analysis and BEs by energy simulation and comparing different materials, yet there is no unified agreement on the most effective envelopes and materials in reducing energy consumption and enhancing thermal comfort. For example, some studies advocate green roofs as being optimal for energy savings, while others find traditional insulation methods more beneficial. Similarly, opinions differ on the best wall materials and insulation. These discrepancies underscore the need for a detailed comparative analysis that covers a wide range of BEs specific to the studied region.

Moreover, many existing studies do not incorporate BIM for detailed building design optimization. In facility management, BIM may significantly contribute to energy management through the automation of energy modeling processes, the augmentation of existing data libraries, and the efficient organization of building data [36]. In contrast, this study aims to develop innovative solutions by utilizing BIM to integrate various building features and optimize design, providing a more thorough approach to energy management.

Additionally, the use of ML has expanded into many areas, including predicting building energy consumption. Despite extensive research into the application of ML in building design, most studies concentrate exclusively on different ML algorithms for energy prediction and performance but do not explore the combined use of BIM models with ML for optimizing energy efficiency. Such investigations are still relatively scarce, often limited to concepts, small-scale experiments, or early testing phases. By integrating ML with BIM, this study offers a more dynamic and interactive approach to optimizing energy performance.

The primary aim of this study was to systematically compare a wide range of different BEs, such as walls, roofs, flooring, windows, and various other building features, including orientation, infiltration rate, lighting efficiency, daylighting, occupancy control, HVAC types, plug load efficiency, and WWR, using BIM, energy analysis, and ML to assess their impact on energy consumption. Systematically, this study adopts a structured and systematic approach, ensuring a thorough and organized consideration of all relevant variables and their interactions. This approach also provides a detailed understanding of how each building feature influences energy performance, ultimately identifying the most effective strategies tailored to Dubai and Riyadh’s specific climatic and geographical conditions. In conclusion, this study contributes to the field by offering actionable insights for enhancing building energy efficiency in hot climates.

3. Materials and Methods

This study employed a quantitative methodology, using an experimental study design based on computer simulations, to assess and optimize the energy performance of residential buildings. This approach allowed for controlling various building features and examining their impact on energy consumption using the proposed case studies. The simulation and energy estimation regions were applied to two major cities, Dubai and Riyadh, in hot climates in the Middle East. Detailed methodology workflow is shown in Figure 1.

3.1. Study Area and Climate

Both cities fall under Energy Code Climate Zone 1B, characterized as hot and dry. Figure 2 illustrates the geographical locations of the modeled buildings in Riyadh and Dubai, along with the weather stations used for collecting climatic data.

Table 1. Weather station details for Riyadh and Dubai.

	Riyadh	Dubai
Weather Station	GBS_06M12_18_213128	GBS_06M12_18_281137
Location	King Abdulaziz District, Riyadh	Dubai, Dubai
Coordinates	Latitude 24.7000, Longitude 46.7333	Latitude 25.2000, Longitude 55.2833
Elevation	627 m	10 m

provides detailed information about these weather stations. The climatic data were collected from weather stations identified through GBS and Insight. While only one weather station was selected per city, these were chosen for their centrality and representativeness of the broader urban climate, ensuring that the data accurately reflect the typical conditions experienced by most buildings in these regions.

Figure 3 presents the selected weather stations’ monthly temperature data, annual wind rose speed, and relative humidity.

Dubai exhibited higher average temperatures throughout the year, with maximum temperatures peaking around 40 °C in the summer months, while minimum temperatures in winter remained above 10 °C. In contrast, Riyadh showed similar high maximum temperatures during the summer, but minimum temperatures could drop significantly in the winter, even approaching 0 °C. This indicates a more consistent warm climate for Dubai, while Riyadh experiences a larger diurnal temperature range and more seasonal variation.

Dubai displayed higher relative humidity levels, with the most common values ranging between 50% and 70%, indicating a more humid climate. In contrast, Riyadh exhibited lower relative humidity levels, falling most of the time below 20%, highlighting a much drier climate than Dubai.

3.2. BIM Modeling

This study utilized BIM to create detailed 3D models of residential buildings in Dubai and Riyadh. The BIM models were developed using Autodesk Revit (Version 25.0.2.419), a widely used software for creating detailed designs. The models included accurate thermal specifications of different BEs, which is essential for conducting a thorough energy performance analysis.

The models were based on a typical residential villa commonly found in the region, as depicted in Figure 4. The BIM models incorporated various materials, including conventional, modern, and green materials for BEs. This variety allowed for evaluating different construction approaches and their impact on energy efficiency.

3.2.1. Key Features Modeled

Different building scenarios were created and examined to evaluate the impact of various construction components on energy efficiency. Each scenario featured unique configurations, allowing for a thorough analysis across a wide range of features that affect energy performance. These scenarios were essential for assessing the effects of each building feature on energy efficiency, which is crucial for ML algorithms to understand the relationship between each feature and energy consumption.

Key building features analyzed included five main BEs such as exterior wall type, window type, flooring system, interior walls, and roofing types, in addition to other features like floor area, building orientation, infiltration rate, lighting efficiency, daylighting and occupancy control, HVAC types, plug load efficiency, and WWR. Specifically, the study included 15 different types of external walls, seven different types of roofs, two types of flooring, five types of windows, and three types of internal walls. These BEs were selected based on their regional prevalence, environmental benefits, and advanced insulation properties. Selecting commonly used materials ensured the study’s applicability to local construction practices. Green materials were selected to align with sustainability goals. Advanced insulation materials like VIPs and Aerogel were included for their superior thermal performance.

Table 2. Details of evaluated roofs.

No	Roof Type	Layers	Thickness	R-Value	Thermal Mass
R1	Concrete Roof	Gypsum Plaster (GP) (20 mm) + Concrete (200 mm) + Cement Plaster (CP) (15 mm)	235	0.24	343.32
R2	Insulated Concrete Roof	GP (20 mm) + Mineral Wool (100 mm) + Reinforced concrete (200 mm) + Bitumen layer (4 mm) + Roofing Tiles (40 mm)	356.5	3.2	396.55
R3	Green Roof	GP (12.5 mm) + Air Gap (30 mm) + VIPs (100 mm) + Reinforced concrete (150 mm) + Waterproof layer + Drainage layer (50 mm) + Filter layer + Soil (150 mm) + Plant layer (150 mm)	642.5	13.98	799.26
R4	Solar Reflective Roof	GP (12.5 mm) + Aerogel Insulation Board (100 mm) + Concrete Deck (200 mm) + CP (15 mm) + Waterproof Membrane + Reflective Elastomeric Coating (5 mm)	332.5	5.4	357.64
R5	Double-Skin Roof	GP (20 mm) + Air Gap (50 mm) + Reinforced Concrete Slab (150 mm) + Roofing Felt + Air Gap (30 mm) + Metal Panel (2 mm)	252	3.37	253.2
R6	SIP High-Performance Roof	GP (12.5 mm) + Air Gap (50 mm) + EPS Foam (100 mm) + Concrete Slab (150 mm) + EPS Foam (100 mm) + Vapor Retarder + Cement Screed (15 mm) + Asphalt Layer (5 mm)	432.5	7.9	277.45
R7	Sloped Green Roof	Green Roof with 30% Slope	642.5	13.98	799.26

,

Table 3. Details of evaluated flooring.

No	Flooring Type	Layers	Thickness	R-Value	Thermal Mass
F1	Wooden Flooring	Wooden Flooring (20 mm) + Vapor Retarder + Plywood, Sheathing (25 mm) + RW (100 mm) + Timber Joist (150 mm)	195	3.34	54.6
F2	Concrete Flooring	Marble Tiles (30 mm) + Cement Screed (20 mm) + Vapor Retarder + Rigid insulation (100 mm) + Damp-proofing + Concrete, Cast in Situ (150 mm)	300	3.03	329.53

,

Table 4. Details of evaluated interior walls.

No	Interior Wall Type	Layers	Thickness	R-Value	Thermal Mass
IW1	Drywall	GP (12.5 mm) + Rockwool Insulation (50 mm) + Metal Stud Layer (75 mm) + GP (12.5 mm)	150	4.5	30.29
IW2	Concrete Block Wall	GP (12.5 mm) + CP (15 mm) + Concrete Block (100 mm) + CP (15 mm) + GP (12.5 mm)	155	0.14	219.63
IW3	Brick Wall	GP (12.5 mm) + CP (15 mm) + Brick (100 mm) + CP (15 mm) + GP (12.5 mm)	155	0.25	198.63

,

Table 5. Details of examined window types.

No	Window Type	Frame Material	R-Value	U-Value
W1	Single glazing SC = 0.8	Wood	0.14	6.7
W2	Double glazing − domestic SC = 0.6	Aluminum	0.35	2.85
W3	Triple glazing − 1/8 in thick − low-E/clear/low-E (e = 0.1) glass	uPVC	0.65	1.53
W4	Reflective double glazing − 1/4 in thick − 14% stainless steel on green glass	Aluminum	0.5	1.98
W5	Low-E triple glazing SC = 0.2	uPVC	0.68	1.45

,

Table 6. Details of other building features.

Features	Details	Values
Area	Various total floor areas ranging from 170 to 370 m2.	6
WWR	Different ratios range from 0% to 95% of the wall.	4
Building Orientation	Various orientations from 0 to 360 degrees.	8
Infiltration Rate	Different rates from 0.17 to 2.0 ACH.	6
HVAC System	Revit defaults to eight different types of HVAC systems.	8
Plug Load Efficiency	Various rates from 0.6 to 2.6 W/sf.	6
Daylighting and Occupancy Controls	Presence or absence of these controls.	4
Lighting Efficiency	Various rates from 0.3 to 1.9 W/sf.	5
Photovoltaic (PV) Roof	The presence or absence of 60% solar panel coverage on the roof.	2

and

Table 7. Details of evaluated exterior walls.

No	Exterior Wall Type	Layers	Thickness	R-Value	Thermal Mass
EW1	Brick Wall	GP (12.5 mm) + Brick (200 mm)	212.5	0.39	271.95
EW2	Insulated Brick Wall	GP (12.5 mm) + RW (50 mm) + Brick (200 mm)	262.5	1.86	279.05
EW3	Concrete Blocks with Stone Cladding	GP (12.5 mm) + Phenolic Foam Insulation (100 mm) + Concrete Blocks (200 mm) + Cement Mortar (20 mm) + Stone Cladding (40 mm)	372.5	5.8	439.33
EW4	Dense Concrete Wall	GP (12.5 mm) + Concrete (200 mm) + CP (15 mm)	227.5	0.22	336.44
EW5	Double Brick Wall with VIP Insulation	GP (12.5 mm) + Brick (100 mm) + VIPs (100 mm) + Brick (100 mm)	312.5	10.33	315.35
EW6	Double Concrete Wall with Air Gap	GP (12.5 mm) + Concrete Blocks (100 mm) + Air Gap (100 mm) + Concrete Blocks (100 mm) + Mortar Bed (20 mm) + Marble (40 mm)	372.5	4.24	436.51
EW7	Fiberglass Insulated Sandwich Panels	GP (12.5 mm) + Concrete (100 mm) + Fiberglass Batt (100 mm) + Concrete (100 mm) + CP (15 mm)	327.5	5.49	339.38
EW8	Cellulose Insulated Sandwich Panels	GP (12.5 mm) + Concrete (100 mm) + Cellulose Insulation (50 mm) + Concrete (100 mm) + CP (15 mm)	277.5	1.51	339.4
EW9	EPS SIPs	GP (12.5 mm) + OSB (15 mm) + EPS (200 mm) + OSB (15 mm) + Cedar Cladding (10 mm)	252.5	6	65.14
EW10	XPS SIPs	GP (12.5 mm) + OSB (15 mm) + XPS (Extruded Polystyrene) (200 mm) + OSB (15 mm) + Cedar Cladding (10 mm)	252.5	7.7	68.66
EW11	Insulating Concrete Form	GP (12.5 mm) + EPS Foam (100 mm) + Reinforced Concrete (100 mm) + EPS Foam (100 mm) + Exterior Finish (15 mm)	327.5	5.85	192.86
EW12	Aerogel Insulated Wood Frame	GP (12.5 mm) + Wood Frame (150 mm) + Cavity Aerogel Insulation (150 mm) + Exterior sheathing (13 mm) + Rendering (15 mm)	190.5	8	72.89
EW13	Solar Reflective Wall	GP (12.5 mm) + Concrete Block (100 mm) + Vapor Retarder +Air (20 mm) + VIP Insulation (100 mm) + Air (20 mm) + Brick (100 mm) + CP (5 mm) + Reflective Elastomeric Coating (5 mm)	362.5	11.95	341.71
EW14	Green Wall	GP (12.5 mm) + Brick (200 mm) + CP (10 mm) + Waterproof Membrane + Cavity (50 mm) + Polypropylene (50 mm) + Cultivated Soil (80 mm) + Vegetation (80 mm)	482.5	2.89	634.19
EW15	Cavity Green Wall	GP (12.5 mm) + Brick (100 mm) +Air (20 mm) + VIP Insulation (100 mm) + Air (20 mm) + Concrete Block (100 mm) + CP (10 mm) + Vapor Retarder + Soil (80 mm) + Vegetation (80 mm)	522.5	12.14	596.17

detail various materials and other building features used for scenario development. Figure 5 details the cross-sections of the modeled exterior walls.

3.2.2. Scenario Development

Initially, the study commenced with 125 scenarios to provide a baseline evaluation of various building features. The selection of scenarios was carefully designed to maximize coverage and ensure comprehensive analysis without any repetition. This approach was implemented by systematically rotating the building features, such as exterior wall type, roof type, and window type, cyclically. By doing so, the scenarios encompass the broadest possible range of feature combinations, capturing the relationships between building features and energy consumption. The cyclic permutation method ensures that each scenario is unique and provides a reliable dataset for ML models to predict EUI accurately. The initial number of scenarios was chosen based on similar practices in the literature, where a smaller, manageable number of scenarios helps establish foundational insights. For instance, in a study involving the sensitivity analysis of BEP, researchers initially used 120 simulations but later increased this number to achieve stable results. This approach illustrates the iterative process of refining the number of scenarios to accurately capture complex relationships [30]. In this context, starting with 125 scenarios allowed for efficient analysis while establishing initial insights. As the study progressed, the dataset was intentionally expanded to 250 scenarios to evaluate the impact of dataset size on model performance and to accurately capture the complex relationships between building features and energy consumption. This strategic expansion enabled a more comprehensive analysis, allowing for the ML models to understand better the intricate dynamics involved.

3.3. Energy Simulation

The BIM models developed in Autodesk Revit were exported to GBS using the Green Building XML (gbXML) format. The energy simulations were executed using GBS and Autodesk Insight, which are integrated with Revit to facilitate precise energy performance analysis. The primary metric for evaluating BEP was EUI, which measures energy consumed per unit area of the building per year (MJ/${\mathrm{m}}^2/$year). This standardized metric allows for comparing the energy efficiency of different building designs. The simulation results were exported as Comma-Separated Values (CSV) files for later ML use.

Figure 6 provides a 3D visualization of the energy models created with Autodesk Revit and Insight. In this representation, the areas highlighted in blue denote the analytical spaces within each building model. These analytical spaces are critical for conducting energy simulations, as they delineate the zones where energy consumption calculations are performed.

3.4. Machine Learning Models

In this study, ML models were employed to predict the EUI of various building scenarios and to analyze the impact of different building features on energy consumption, forming the basis for later optimization.

3.4.1. Data Preparation

Data preprocessing entails cleaning and converting raw data into a format ready for analysis. As part of this process, to ensure that the features contribute equally to the model training process and to improve the performance, the data were standardized using Standard Scaling with ${\mathrm{x}}_{\mathrm{S}\mathrm{c}\mathrm{a}\mathrm{l}\mathrm{e}\mathrm{d}}={\displaystyle \frac{\mathrm{x}-\mathrm{\mu }}{\mathsf{\alpha }}}$ and with mean $\mu ={\displaystyle \frac{1}{N}} \sum _{i=1}^N\left(x_i\right)$ and standard deviation $\sigma =\sqrt{{\displaystyle \frac{1}{N}} \sum _{i=1}^N{(x_i-\mu )}^2}$, where x is the original feature value, μ is the mean of the feature values, and σ is the standard deviation of the feature values. This transformation adjusts the data with a mean of zero and a standard deviation of one. The μ and σ values for each BE, along with their corresponding normal distribution plot, are illustrated in Figure 7. Standard scaling was selected because it is beneficial for SVM and LR, which are sensitive to the scale of the input data. The same scaling method was applied to all models to ensure consistency across the analysis.

The dataset was split into training and testing sets using an 80/20 split ratio to ensure an unbiased evaluation of the model’s performance and generalization to unseen data. This ratio was chosen based on the specific needs of the model and dataset. Recent studies suggest that the optimal training/testing split ratio can be approximated using the formula $\mathsf{\rho }=\sqrt{\mathrm{p}}:1$, where p is the number of the model’s features. With 14 features in this case, the 80/20 ratio aligns closely with this optimal distribution, providing enough data for effective model training while allowing for an accurate assessment of its generalization capabilities [37].

To further enhance the evaluation and robustness of the models, 10-fold cross-validation was employed. In this method, the training data were divided into ten equal-sized folds. The model was then trained and validated ten times, each time using a different fold as the validation set and the remaining nine folds as the training set, as shown in Figure 8. This approach ensures that every data point has the opportunity to be used in both training and validation. Using 10-fold cross-validation offers a balance between computational efficiency and thorough evaluation. By averaging the performance metrics across the ten iterations, a more reliable estimate of the model’s effectiveness is obtained.

GridSearchCV was employed to optimize the performance of the ML models for hyperparameter tuning. For the SVM, the parameters tuned included the kernel type (‘kernel’), regularization parameter (‘C’), and kernel coefficient (‘gamma’). For the RF, the parameters tuned included the number of trees (‘n_estimators’) and maximum depth (‘max_depth’). For the GBM model, using the ‘GradientBoostingRegressor’ class from scikit-learn, GridSearchCV was used to optimize hyperparameters such as learning rate (‘learning_rate’), number of estimators (‘n_estimators’), and maximum depth (‘max_depth’). By systematically searching through these parameter grids and applying 10-fold cross-validation within each combination, GridSearchCV identifies the optimal set of hyperparameters, enhancing the model’s performance and ensuring reliable results.

Feature analysis was conducted to understand the importance and contribution of each feature to the model’s performance. Various techniques were used, including feature importance from tree-based models, Pearson’s correlation coefficient (PCC), and SHAP (SHapley Additive exPlanations).

In tree-based models, feature importance scores were derived, indicating each feature’s relative importance in making predictions. Features with higher importance scores influence the model’s output more. PCC was used to measure the linear correlation between each feature and the target variable (EUI). This statistical measure helps identify features with a strong linear relationship with the target or each other, and it provides insights into which features are likely to be more predictive. SHAP values were employed to provide a more complex understanding of BEs’ contribution to create a baseline for further optimization. SHAP considers both the magnitude and direction of each BE effect on the prediction, enhancing model interpretability and transparency.

3.4.2. Model Training

The following ML models were selected for the analysis due to their effectiveness in regression tasks and their ability to handle complex datasets, as supported by the literature; however, this study specifically explored their effectiveness in this research context.

LR is a linear regression method that incorporates a regularization parameter to mitigate overfitting. It penalizes the absolute values of the regression coefficients, effectively reducing some of them to zero. LR helps reduce overfitting and is particularly useful when several independent variables may be irrelevant. Studies have shown that LR is effective for feature selection in high-dimensional datasets, making it suitable for this study where multiple building features are analyzed simultaneously [38].

$${\widehat{\mathsf{\beta }}}^{\mathrm{L}\mathrm{a}\mathrm{s}\mathrm{s}\mathrm{o}}={\mathrm{m}\mathrm{i}\mathrm{n}}_{\mathsf{\beta }}\left({\displaystyle \frac{1}{2\mathrm{N}}}{\sum }_{\mathrm{i}=1}^{\mathrm{N}}{\left({\mathrm{E}\mathrm{U}\mathrm{I}}_i-{\sum }_{\mathrm{j}=1}^{\mathrm{P}}{\mathrm{F}\mathrm{e}\mathrm{a}\mathrm{t}\mathrm{u}\mathrm{r}\mathrm{e}}_{\mathrm{i}\mathrm{j}}{\mathsf{\beta }}_{\mathrm{j}}\right)}^2+\mathsf{\lambda }{\sum }_{\mathrm{j}=1}^{\mathrm{P}}\left|{\mathsf{\beta }}_{\mathrm{j}}\right|\right)$$

where

• ${\mathrm{E}\mathrm{U}\mathrm{I}}_i$ is the observed Energy Use Intensity for the i-th scenario,
• ${\mathrm{F}\mathrm{e}\mathrm{a}\mathrm{t}\mathrm{u}\mathrm{r}\mathrm{e}}_{\mathrm{i}\mathrm{j}}$ is the value of the j-th building feature in the i-th scenario,
• ${\mathsf{\beta }}_{\mathrm{j}}$ are the regression coefficients for the building features,
• $\mathsf{\lambda }$ is the regularization parameter,
• $\mathrm{N}$ is the number of scenarios (building configurations),
• $\mathrm{P}$ is the number of building features.

SVM is a supervised learning model employed for regression analysis. It aims to fit the optimal hyperplane within a predefined margin of tolerance. Unlike traditional regression methods, SVM introduces the concept of an epsilon (ε)-insensitive margin around the regression line, allowing for deviations within a specified margin. SVMs are known for handling non-linear relationships using kernel functions, which are crucial for capturing the complex interactions in the data [39]. For non-linear SVMs, the decision function is defined using a kernel function $K (x_i,x_j)$ that transforms the input data into a higher-dimensional space. This transformation allows the model to capture the complex, non-linear interactions between the building features and energy consumption, as depicted in Figure 9.

RF is an ensemble learning algorithm that generates numerous decision trees during the training phase and produces the average prediction derived from these trees (refer to Figure 10). This algorithm is renowned for its accuracy and ability to manage large datasets with high dimensionality. Its robustness against overfitting and capability to model complex relationships make it a popular choice in energy prediction studies [40].

GBM is a powerful ensemble learning method that sequentially constructs trees, with each new tree addressing the mistakes made by the preceding trees, making the model more robust and accurate over iterations. GBM has been shown to provide superior predictive performance by combining the strengths of multiple weak learners, and it is particularly effective in regression tasks where capturing subtle patterns is necessary [41].

3.4.3. Model Evaluation

Model performance is quantitatively assessed using the below metrics, which provides a comprehensive evaluation scheme that reflects the accuracy of the predictive models.

Mean Absolute Error (MAE) provides a linear scale of the errors in prediction, representing the average magnitude of the errors in a set of predictions without considering their direction (positive/negative).

$$\mathrm{M}\mathrm{A}\mathrm{E}={\displaystyle \frac{1}{\mathrm{n}}}\sum _{\mathrm{i}=1}^{\mathrm{n}}\left|{\mathrm{y}}_{\mathrm{i}}-{\widehat{\mathrm{y}}}_{\mathrm{i}}\right|$$

where $y_i$ is the actual value, $\widehat{y_i}$ is the predicted value from the model, and n is the number of observations.

Root Mean Squared Error (RMSE) is the square root of the mean of the squared errors. RMSE is particularly sensitive to large errors and, thus, effectively captures the variability in data.

$$\mathrm{R}\mathrm{M}\mathrm{S}\mathrm{E}=\sqrt{{\displaystyle \frac{1}{\mathrm{n}}}\sum _{\mathrm{i}=1}^{\mathrm{n}}{\left({\mathrm{y}}_{\mathrm{i}}-{\widehat{\mathrm{y}}}_{\mathrm{i}}\right)}^2}$$

R-squared (R²) indicates how much of the variability in the dependent variable can be explained by the independent variables, offering an understanding of the model’s fit quality.

$${\mathrm{R}}^2=1-{\displaystyle \frac{\sum _{\mathrm{i}=1}^{\mathrm{n}}{\left({\mathrm{y}}_{\mathrm{i}}-{\widehat{\mathrm{y}}}_{\mathrm{i}}\right)}^2}{\sum _{\mathrm{i}=1}^{\mathrm{n}}{\left({\mathrm{y}}_{\mathrm{i}}-{\overline{\mathrm{y}}}_{\mathrm{i}}\right)}^2}}$$

where $\overline{y}$ is the mean of the observed data.

3.5. Optimization

The final step in the methodology involved optimizing the building scenarios to identify the most energy-efficient configurations. The optimization process began with an initial analysis of the simulation results and the FIA conducted during the ML phase, which provided a valuable understanding of which building features had the most impact on energy consumption.

The GBM model was specifically used to predict the top ten Best-Case Scenarios (BCS) and the worst-case scenario (WCS) from all potential combinations of BEs. Generated optimized building designs were validated through additional manual BIM and energy simulations to ensure their effectiveness. Predicted EUI values were then compared with the actual results of the additional simulations to validate the model performance. Additionally, the EUI values of these optimized designs were compared to the initial scenarios to quantify the improvements achieved through optimization.

4. Results

This section presents a detailed analysis of the performance of four ML algorithms—LR, SVM, RF, and GBM—in predicting EUI for various building scenarios in DS and RS. The results also include an evaluation of FIA, the most essential building features and optimization, and the analysis of BCS and WCS.

4.1. Model Performance Analysis

The dataset comprised two stages of analysis, initially starting with 125 scenarios and subsequently expanded to 250 scenarios. The average performance metrics for each model in both the initial and expanded datasets are detailed in

Table 8. Evaluation results for ML models.

	GBM		RF		SVM		LR
Evaluation Results	Initial Dataset	Expanded Dataset	Initial Dataset	Expanded Dataset	Initial Dataset	Expanded Dataset	Initial Dataset	Expanded Dataset
MAE	9.53	7.06	22.98	14.92	33.39	70.1	56.07	73.6
RMSE	18.1	11.29	44.87	25.15	57.29	98.41	73.55	104.25
R2	0.923	0.989	0.707	0.942	0.346	0.468	0.386	0.593

and Figure 11.

According to the literature, an R² value exceeding 0.8 in ML models for predicting energy consumption is a robust indicator of model reliability and accuracy [42], and values closer to 1 indicate a higher accuracy and predictive power.

In the initial dataset analysis, the GBM model demonstrated superior performance compared to the other models. It achieved the lowest error rates and the highest R² values of 0.923, indicating a strong fit to the data and an ability to capture complex relationships within the dataset. The RF model showed moderate performance with better results than LR and SVM, but it still exhibited higher error rates and a lower R² value of 0.707 compared to GBM. Both LR and SVM models struggled with the complexity of the dataset. They produced the highest error rates and lowest R² values of 0.386 and 0.346, respectively, indicating poor performance and a lack of predictive power.

Expanding the dataset yielded significant improvements in model performance. The additional data allowed the models to capture better the complex relationships between building features and energy consumption, leading to more accurate predictions. Specifically, the GBM model’s R² value increased to 0.989, indicating a 7.1% improvement in its ability to explain the variance in EUI. Similarly, RF showed a 33.2% increase in R² value, from 0.707 to 0.942, reflecting its enhanced predictive accuracy. Both models exhibited substantial reductions in MAE and RMSE values, demonstrating their improved fit to the data. Although the SVM and LR models benefited from the expanded dataset, their performance gains were relatively modest. SVM’s R² value increased from 0.346 to 0.468, a 35.3% improvement, and LR’s R² value increased from 0.386 to 0.593, a 53.6% improvement, but they still lagged behind GBM and RF in terms of predictive power and accuracy.

The GBM model’s performance improved with the doubling of the data, but it was less affected by the data size increase compared to other models, indicating its superior performance regardless of dataset size. Overall, the expanded dataset enabled a more robust and reliable model performance, highlighting the importance of dataset size in ML-based energy consumption prediction.

4.1.1. Visual Assessments of Model Predictions

Scatter plots (refer to Figure 12 and Figure 13) were generated to visually assess the accuracy of the EUI predictions. These plots compare the actual EUI values from the dataset against the predicted EUI generated by the ML models. Each plot includes a red line representing a perfect prediction accuracy, where predicted values match the actual values exactly. By analyzing the clustering of data points around this line, it becomes evident which models are more reliable and accurate for EUI prediction [43]. This visual assessment reinforces the quantitative metrics discussed earlier.

The scatter plot for the GBM model demonstrates a high degree of accuracy in EUI predictions, with data points clustering closely around the line, representing a perfect prediction accuracy. This tight clustering indicates that the GBM is highly effective at capturing and predicting the complex patterns of energy use within the buildings studied.

The scatter plot for RF shows a satisfying clustering of points near the line of perfect prediction, underscoring the model’s capability to predict the EUI accurately. However, the scatter plots for the SVM and LR models reveal a more dispersed set of points, straying farther from the line of perfect prediction.

Figure 14 and Figure 15 are the residual plots, mapping residuals on the y-axis against predicted values on the x-axis. Ideally, residuals should be randomly dispersed around the horizontal line without displaying systematic patterns such as funnels, curves, or trends. Random dispersion indicates that the model’s predictions are unbiased and accurate [44].

In this analysis, the RF and GBM models demonstrate a random dispersion of residuals closely clustered around the zero line, confirming the reliability of these models. In contrast, the SVM and LR models exhibited a significantly weaker performance. The residuals for these models were more widely dispersed and deviated from the zero line, indicating less precision in their predictions.

4.1.2. Comparison with the Existing Literature

While a significant body of literature supports the accuracy of LR in predicting energy consumption, some studies align with the results of this research, indicating weaknesses in the LR model. For instance, Bahij et al. [45,46] highlighted the efficiency of LR in specific scenarios. However, these findings are contrasted by other studies that demonstrate the LR model’s inadequacy in capturing complex relationships within energy consumption data. Specifically, Archibong-Eso et al. [47] noted that SVM outperforms LR due to its superior ability to handle nonlinear relationships and complex interactions among variables.

Conversely, the performance of SVM in predicting energy performance has been well documented. Seyedzadeh et al. [34] showed the potential of SVM models in accurately forecasting energy usage. Despite this, traditional SVM approaches have been criticized for their struggles with high nonlinearity between inputs and outputs in building energy consumption models. Zhong et al. [48] argued that such limitations necessitate the use of advanced techniques like vector field-based methods to enhance accuracy and robustness. Moreover, Yang et al. [49] found that improved SVM models, such as those optimized with the Sparrow Search Algorithm, significantly outperformed traditional SVM, underscoring the necessity for optimization to achieve a high prediction accuracy.

These findings suggest that while both LR and SVM models have their strengths, their effectiveness can vary significantly based on the complexity of the data and the specific application. Although straightforward and easy to implement, LR often falls short in scenarios involving intricate relationships. In contrast, SVM, while more robust in handling nonlinear data, requires careful tuning and optimization to fully realize its predictive capabilities.

The recent literature strongly supports the superior performance of the GBM in predicting energy consumption. Papadopoulos et al. [50] evaluated tree-based ensemble learning algorithms, including GBM, for building energy performance estimation and found that GBM significantly improved prediction accuracy. Similarly, Barbaresi et al. [35] investigated various ML models for predicting building energy needs and reported that Extreme Gradient Boosting (XGBoost), a variant of GBM, showed the best performance among all tested models. Additionally, Han et al. [51] applied XGBoost to predict energy use in construction and demonstrated that the model achieved a high accuracy, outperforming traditional statistical methods and other ML algorithms.

4.2. Feature Importance Analysis

The variables that influence energy efficiency in buildings in DS and RS are examined in Figure 16 and Figure 17 through the FIA conducted using GBM. This comparative analysis reveals similarities and distinctions in how various features influence energy consumption in these two cities’ different simulated building scenarios.

In contrast to DS, RS places more emphasis on the roof and external walls, indicating the necessity for appropriate insulation and construction materials in harsher climates. DS exhibits a balanced distribution of importance among various features.

Roofs emerged as the most influential feature in both DS and RS, accounting for 29% and 40% of the overall importance, respectively. This substantial impact is attributed to the roof’s role as the primary barrier against solar radiation in hot climates. The roof’s thermal performance becomes critically important in regions like Riyadh, where the climate is characterized by intense solar exposure and higher temperature extremes. A roof’s ability to reflect solar radiation and minimize heat absorption directly influences the internal cooling load, making it a decisive factor in a building’s overall energy efficiency. The higher percentage in Riyadh underscores the increased significance of roof insulation and reflective materials in regions with more extreme climatic conditions, where roof performance can drastically affect energy consumption.

Exterior walls also play a crucial role, contributing 19% in DS and 29% in RS. The thermal properties of exterior walls are vital for controlling heat transfer between the building’s interior and exterior environments. Well-insulated exterior walls are particularly critical in maintaining internal temperature stability and reducing reliance on HVAC systems. The higher contribution of exterior walls in Riyadh suggests that in climates with larger diurnal temperature variations, the insulating capacity of the exterior walls is a crucial determinant of energy efficiency. Effective wall insulation reduces the heat flux through the building envelope, helping to maintain comfortable indoor conditions with lower energy input.

Windows plays a vital role in both datasets, contributing 15% in DS and 9% in RS. The higher importance of DS suggests that a greater focus on optimizing natural lighting and reducing cooling loads through efficient window design can highly impact energy consumption.

Plug load efficiency, at 9% in DS and 5% in RS, emphasizes the importance of energy-efficient appliances and electronic devices in reducing overall energy consumption. This feature’s consistent importance across both datasets highlights the universal need for efficient plug loads.

The presence of PV roofs can contribute 7% in DS and 4% in RS. This reflects the growing importance of integrating renewable energy sources to enhance energy efficiency. Having the right HVAC system is crucial for maintaining indoor thermal comfort and operational efficiency, contributing 7% in DS and 3% in RS.

WWR affects energy performance by balancing natural light and heat transfer, with contributions of 7% in DS and 4% in RS. Lighting efficiency, accounting for 5% in DS and 3% in RS, underscores the role of advanced lighting technologies in reducing energy consumption. Efficient lighting contributes to energy savings.

The radar chart (Figure 18) visually represents the feature importance values for DS and RS in a logarithmic scale base 10, allowing for a better comparison of the influential factors in both datasets. This chart highlights that area, infiltration rate, flooring type, daylighting and occupancy controls, interior wall type, and building orientation have minimal individual impacts; however, they collectively contribute to the overall energy efficiency. The importance of these features varies slightly between the two datasets, as distinct climatic conditions and building characteristics influence them.

The heat map in Figure 19 illustrates the PCC between building features and EUI. Ranging from −1 to 1, PCC quantifies the degree of linear correlation between two variables, with larger absolute values indicating stronger correlations.

The differences observed between PCC and FIA can be attributed to the nature of each metric. PCC captures direct linear relationships, showing higher values for features like windows, flooring, interior walls, and area, indicating their potential for energy efficiency improvements. However, their lower FIA scores suggest their impact is more complex, considering interactions with other features. Conversely, features like building orientation, daylighting and occupancy controls, infiltration rate, roof, and exterior wall show similar results in both metrics, underscoring their roles in energy consumption through clear linear relationships and combined effects within the model. Roof and exterior walls, in particular, are consistently substantial. This highlights the need to consider both linear relationships and complex interactions in developing detailed energy efficiency strategies.

The SHAP summary plot (refer to Figure 20) presents a detailed visualization of the effect of different BEs on the predictions made by GBM. Each BE is listed on the Y-axis, ordered by its importance, with the most impactful ones at the top. The SHAP values, displayed along the horizontal line, demonstrate the effect of each BE on the model’s output, in which negative values reflect a decrease, whereas positive values signify an increase in the prediction. The color coding of the dots ranges from blue (low values) to red (high values), showing how the value of each BE influences the prediction. This plot helps understand the magnitude and direction of each BE’s impact on energy consumption predictions.

The SHAP summary plot reveals that walls and roofs are the most significant BEs influencing energy consumption predictions, with walls showing both positive and negative impacts. Higher values of the wall (red dots) generally decrease energy consumption, while lower values (blue dots) tend to increase it. Similarly, the roof exhibits a varied impact, with high values leading to reduced energy consumption. Windows also plays a crucial role, where higher values significantly elevate the predictions. Internal walls and floors have a lesser but noticeable impact, with higher values of internal walls generally decreasing energy consumption and floors showing a more spread-out influence.

4.3. Building Envelopes Energy Analysis

This section explores BEs and their impact on energy consumption. The significance of each building feature is meticulously quantified. Features such as plug or lighting load, infiltration rate, building area, and WWR have clear implications; overall energy consumption correspondingly diminishes as these factors decrease. Adhering to each country’s regulatory frameworks further facilitates applying these insights.

However, the role of BEs stands out as particularly crucial in influencing energy performance. According to the FIA, BEs are responsible for approximately 70% of the energy consumption in DS and over 84% in RS. Specifically, three types of BEs—roofs, exterior walls, and windows—account for over 62% of DS and 78% of RS. This significant impact underscores the need for strategic design and material selection to optimize energy efficiency. These three types of BEs were analyzed through additional simulations with GBS, with all variables kept similar to the base scenario (BS) while only the BE under study was varied. The BS is a scenario created using the following configuration: EW1, R1, W1, IW1, F1, an area of 170, a WWR of 30%, and other default settings. The purpose of this scenario is to serve as a baseline for comparison, allowing for the effects of varying individual BEs to be accurately assessed. The results are presented in the following subsections.

4.3.1. Analysis of Roof Types

Roof types exert the highest influence on energy consumption within the BEs. The bar chart in Figure 21 presents a comparative analysis of the EUI for different simulated roofs.

The results indicate that R7 (sloped green roof) is the most energy-efficient option in both datasets. In RS, this roof type can reduce energy consumption by up to 24% compared to BS, while in DS, the reduction potential is approximately 17%. These energy savings are attributable to the R7 excellent insulation properties, and the cooling effect provided by the vegetation layer. Additionally, it is noteworthy that R7, compared to R3 (flat green roof), can reduce energy consumption by up to 3.2%.

R6 (SIPs high-performance roof) is identified as the next most energy-efficient type, potentially reducing energy consumption by more than 21% in RS and up to 14% in DS.

Conversely, R1 (concrete roof) exhibits the poorest performance among all evaluated roof types in both datasets. The subpar performance of R1 is due to its high thermal conductivity, which facilitates heat gain and consequently increases the cooling load.

4.3.2. Analysis of Exterior Wall Types

External walls were marked as the second highest influence on energy consumption within the BEs. The bar chart in Figure 22 presents a comparative analysis of the EUI for different simulated walls.

The results clearly show that EW15 (green cavity wall) is the most energy-efficient type of wall among both datasets. In RS, this wall has the potential to reduce energy consumption by up to 27% compared to the BS and up to 17% in DS. This reduction can be attributed to the superior insulation properties and thermal mass provided by EW15, which effectively reduces heat transfer into the buildings, thereby lowering the cooling load.

The second most energy-efficient wall in RS is EW13 (solar reflective wall), which can reduce energy consumption by up to 24%. This performance is due to the reflective properties of the wall, which minimizes the absorption of solar radiation, reducing the cooling demand. In DS, EW5 (double brick wall with VIPs) can decrease the building’s energy use by 14%. The inclusion of VIPs in the wall construction provides excellent thermal resistance, which is crucial in hot climates.

The results clearly indicate that EW4 (dense concrete wall) performs the poorest among all evaluated walls, increasing energy consumption by up to 9% in RS and more than 5% in DS compared to BS. This poor performance is due to the high thermal conductivity of dense concrete, which allows more heat to penetrate the BE, increasing the need for cooling.

Additionally, the bar chart reveals that similar to the roofs, RS generally exhibits better energy performance than DS, as almost all wall types show lower energy consumption in RS than in DS, except for the brick wall and dense concrete wall.

4.3.3. Analysis of Window Types

Windows were identified as the third highest influence on energy consumption within the BE. The bar chart in Figure 23 provides a comparative analysis of the EUI for different simulated window types.

The results indicate that W5 (low-E triple glazing windows) is the most energy-efficient in RS, potentially reducing energy consumption by up to 4.4% compared to the BS. In DS, W4 (reflective double-glazing windows) exhibit the best performance, offering up to a 7% reduction in energy consumption. These types of windows are effective in minimizing heat transfer and reducing cooling loads due to their advanced insulating properties.

Conversely, W1 (single-glazed windows) performs the worst among all evaluated window types, significantly increasing energy consumption due to their poor insulation properties.

Additionally, the bar chart reveals that DS generally exhibits better energy performance than RS for all window types. This difference may be attributed to variations in climatic conditions between the two cities.

4.3.4. Comparison with the Existing Literature

The analysis underscores the critical role of BEs in influencing energy performance, with BEs accounting for significant portions of energy consumption in both DS and RS. Specifically, roofs, exterior walls, and windows contribute substantially to overall energy use. This is reinforced by the study on determining optimum insulation thicknesses for external walls and roofs, which emphasizes that insulating these components is the most cost-effective way to control outside elements and reduce fuel consumption and operational costs, despite increased investment costs [52].

Further supporting these findings, recent studies have emphasized the importance of enhancing insulation and selecting appropriate materials for roofs and external walls. Research indicates that enhancing insulation in outer walls and roofs markedly reduces cooling loads in hot climates [16]. Similarly, other studies have demonstrated that thermal mass walls and outdoor green spaces significantly diminish cooling energy demands [18]. These findings are corroborated by the current study, highlighting the importance of selecting appropriate materials and designs to improve energy efficiency.

The study also validates the effectiveness of green roofs and advanced insulation techniques, aligning with the conclusions of Ragab and Abdelrady [25], who observed that green roofs substantially reduce cooling energy requirements in Egyptian school buildings. Alyami et al. [17] illustrated that various green building strategies, including green roofs, effectively achieve ZEB in Saudi Arabia’s hot climate. This analysis identifies the sloped green roof as the most energy-efficient option, significantly lowering energy consumption. This finding underscores green roofs’ excellent insulation properties and cooling benefits, demonstrating their viability for enhancing energy efficiency in hot climates. This is further supported by simulations of the energy performance of buildings with green roofs and green walls in tropical climates, which demonstrated significant decreases in total energy consumption and cooling demand, further reinforcing the benefits of these green building strategies [27].

In summary, the results of this study are in harmony with the existing literature, reinforcing the importance of selecting suitable BEs and insulation materials for optimizing BEP in hot climates. The detailed evaluation of different BE configurations provides valuable insights into their impact on energy consumption, offering practical guidance for designing energy-efficient buildings in hot environments.

4.4. Machine Learning-Based Optimization

This section focuses on the optimization of energy performance using ML. For this purpose, the developed GBM model was employed to predict the top ten BCS and the WCS from all potential combinations of BEs.

These predicted scenarios were then modeled using Revit BIM and subjected to energy simulations with GBS. The results of these simulations were compared with the ML predictions to assess the accuracy and effectiveness of the GBM model in optimizing energy performance.

Table 9. Top ten suggested material-based scenarios with predicted and actual EUI for RS.

No	Exterior Wall	Roof	Window	Flooring	Interior Wall	Predicted EUI	Actual EUI
1	EW15	R7	W4	F2	IW1	430.8	432.8
2	EW15	R7	W4	F1	IW1	431.6	433.5
3	EW15	R7	W4	F1	IW2	432.1	434.6
4	EW15	R7	W4	F2	IW2	432.4	433.9
5	EW15	R7	W5	F2	IW1	433.9	439.8
6	EW13	R7	W4	F1	IW2	434.2	434.3
7	EW13	R7	W4	F1	IW1	434.2	433.2
8	EW15	R7	W5	F1	IW1	434.7	440.5
9	EW15	R7	W5	F1	IW2	434.9	441.6
10	EW13	R7	W4	F1	IW3	435.0	434.0

and

Table 10. Top ten suggested material-based scenarios with predicted and actual EUI for DS.

No	Exterior Wall	Roof	Window	Flooring	Interior Wall	Predicted EUI	Actual EUI
1	EW5	R7	W4	F2	IW1	465.8	466.1
2	EW5	R7	W4	F1	IW2	466.7	468.8
3	EW5	R7	W4	F1	IW3	467.1	468.5
4	EW5	R7	W4	F1	IW1	468.0	467.3
5	EW5	R7	W4	F2	IW2	468.6	467.7
6	EW5	R7	W4	F2	IW3	468.7	467.3
7	EW15	R7	W4	F1	IW1	468.8	470.4
8	EW7	R7	W4	F2	IW1	468.9	468.8
9	EW15	R7	W4	F2	IW1	469.0	469.2
10	EW7	R7	W4	F1	IW1	469.0	469.9

present the details of the top ten BE combinations identified by the GBM model and the actual data derived from the energy analysis for RS and DS.

The predictions made by the GBM model closely align with the actual EUI values, again validating its reliability and effectiveness in predicting energy consumption. The MAE of the predicted scenarios is impressively low, at 2.83 for the RS and 0.96 for the DS, underscoring the model’s accuracy and dependability.

4.4.1. The Worst Scenario for the Case Study Located in Both Riyadh and Dubai

The WCS exhibited the lowest energy performance in both RS and DS. This scenario featured EW4 with 200 mm concrete covered by 15 mm CP and R1 consisting of a concrete structure with a 200 mm concrete core. Figure 24 presents the BIM model of this scenario with corresponding BEs, and the energy simulation results are shown in

Table 11. Energy analysis results of the WCS.

Location	$\mathbf{EUI} (\mathbf{MJ}/{\mathbf{m}}^2/$Year)	Electricity Use (kWh)	Fuel Use (kWh)	Total Energy (kWh)
Riyadh	894.7	70,231	18,116	88,346
Dubai	800.7	75,813	3249	79,061

.

BEs show poor energy efficiency characteristics due to low R-values and high U-values, indicating inadequate insulation. For instance, EW4 and R1, with R-values of 0.22 and 0.24, respectively, and U-values of 4.45 and 4.23, suggest significant heat transfer. This results in higher energy requirements for maintaining indoor temperatures. Also, W1, with a high U-value of 6.7, further contributes to energy inefficiency by allowing for substantial heat loss or gain.

The EUI of Riyadh’s worst-case scenario (RWCS) is alarmingly high at 894.7 MJ/m²/year. This high demand is driven by the extreme temperatures, which necessitate substantial energy for both heating and cooling. The total annual energy consumption stands at 88,346 kWh, with a significant portion allocated to electricity use (70,231 kWh) and a considerable amount to fuel use (18,116 kWh). The substantial fuel use reflects the need for extensive heating solutions and cooling.

The EUI of Dubai’s worst-case scenario, while slightly lower than RWCS, is still high at 800.7 MJ/m²/year. This reflects the hot climate, which drives substantial cooling demands. The total annual energy consumption is 79,061 kWh, predominantly from electricity use (75,813 kWh) and minimal fuel use (3249 kWh). This indicates a heavier reliance on electricity for cooling purposes.

4.4.2. The Best Scenario for the Case Study Located in Riyadh

In Riyadh’s Best-Case Scenario (RBCS), the predicted EUI values range from 430.8 to 435.0, while the actual EUI values range from 432.8 to 441.6. The close alignment between predicted and actual EUI indicates that the GBM model performs well in forecasting the energy performance of RS buildings.

The RBCS, with the lowest predicted EUI and highest energy efficiency, includes an EW15, consisting of two layers of brick and concrete block with an intervening cavity insulated with VIPs, covered with soil and plant layers. It also features R7, a concrete structure insulated with VIP and covered with 150 mm of soil and plant vegetation layers. RBCS also uses W4, concrete flooring (F2), and drywall (IW1). This combination yields a predicted EUI of 430.8 and an actual EUI of 432.8, demonstrating the model’s high accuracy in identifying efficient combinations of building features. The RBCS’ BIM model and corresponding BEs are presented in Figure 25, and the energy simulation results are shown in

Table 12. Energy analysis results of RBCS and DBCS.

Location	$\mathbf{EUI} (\mathbf{MJ}/{\mathbf{m}}^2/$Year)	Electricity Use (kWh)	Fuel Use (kWh)	Total Energy (kWh)
Riyadh	432.8	40,599	3501	44,100
Dubai	466.1	44,337	3233	47,570

. Notably, scenarios with cavity-green exterior walls and sloped green roofs consistently appear among the top performers, highlighting the significant impact of these features on energy efficiency in Riyadh.

This combination demonstrates a significant improvement in energy efficiency compared to the WCS. The substantial reduction in EUI to 432.8 MJ/m²/year highlights the effectiveness of the green envelopes utilized in this scenario. Using VIP-insulated cavity walls with a vegetation layer contributes to superior thermal performance, drastically reducing heat ingress. R7, with its multiple insulating layers and vegetation, further enhances the building’s ability to maintain indoor thermal comfort, thus lowering cooling demands.

4.4.3. The Best Scenario for the Case Study Located in Dubai’s

In Dubai’s Best-Case Scenario (DBCS), the predicted EUI values range from 465.8 to 469.0, while the actual EUI ranges from 466.1 to 470.4. Like RBCS, the GBM model shows solid predictive performance with an MAE of approximately 0.96 EUI, indicating better accuracy than RBCS.

The DBCS, with the lowest predicted EUI, features EW5 and the same BEs as the RBCS. This combination results in a predicted EUI of 465.8 and an actual EUI of 466.1. This close match further validates the model’s effectiveness in predicting energy-efficient combinations. The DBCS’ BIM model and its BEs are provided in Figure 26, and the corresponding energy simulation results are displayed in Table 12.

A notable observation is the frequent appearance of double brick exterior walls with VIP insulation and sloped green roofs among the top scenarios, underscoring their importance in achieving optimal energy performance in Dubai’s climate.

This combination also shows a marked improvement in energy performance. The EUI is reduced to 466.1 MJ/m²/year, reflecting the efficiency gains from the advanced BE design. The double brick walls, equipped with VIP insulation, and the extensive green roof with soil and vegetation layers provide excellent thermal regulation, significantly cutting down on the cooling load. The higher electricity use, at 44,539 kWh, corresponds to the cooling-dominated energy demands of Dubai’s hot climate. Despite the slightly higher EUI compared to RBCS, this scenario still represents a substantial enhancement in energy efficiency.

5. Discussion

5.1. The Novelty of the Study

This study stands out for its innovative integration of BIM and ML to optimize energy performance in residential buildings in hot climates. The research achieves dynamic and accurate simulations of various building scenarios by combining BIM’s detailed digital models with ML’s predictive power. This methodological approach and a detailed analysis of BEs and features provide a holistic understanding of energy consumption factors, significantly advancing current knowledge in the field.

Moreover, the study’s region-specific focus on the hot climates of Dubai and Riyadh addresses a critical gap in existing research, offering tailored and actionable insights for enhancing energy efficiency in similar environments. Detailed BIM models and extensive energy simulations using GBS and Insight ensure the accuracy and reliability of the findings. Additionally, the FIA and subsequent optimization process identify key building features that significantly impact energy consumption, guiding the development of practical recommendations for sustainable building design. This dual approach of prediction and optimization, supported by an iterative expansion of scenarios, underscores the study’s novelty and practical relevance, making it a valuable contribution to the literature on sustainable building practices and energy management.

5.2. The Value of the Findings

The findings of this study offer valuable contributions to academic research and practical applications in sustainable building design, particularly for hot climates. The energy efficiency recommendations derived from analyzing various building features provide practical guidance for architects, engineers, and builders. This facilitates the design of residential buildings optimized for hot climates, ultimately reducing energy consumption and operational costs.

The study’s data-driven approach also supports informed decision-making by utilizing advanced simulation tools and predictive models. This ensures stakeholders can evaluate different design scenarios and make choices that enhance energy efficiency. By resolving discrepancies in the existing literature regarding the most effective BEs and materials, the research offers a clear understanding of the strategies that best contribute to energy savings and thermal comfort. This holistic approach advances academic knowledge and provides actionable insights that can be directly implemented in the construction industry, promoting sustainability and cost efficiency.

5.3. Theoretical, Practical, and Policy Implications

This study contributes to the body of knowledge in three ways. First, it presents an experimental example that expands the use of BIM and ML together in building energy optimization. This integration has proven to be effective; the GBM model consistently outperformed other models, demonstrating superior accuracy and robustness in predicting EUI, with an R² value of 0.99 (refer to Section 3.1, Model Performance Analysis, for detailed results). Second, by systematically comparing a wide range of BEs and features, the study promotes the understanding of how different building features collectively impact energy efficiency. This contrasts with previous studies that often focused on isolated variables, thus encouraging a more integrated perspective in future research. Third, the FIA provides insights into which building features most significantly influence energy consumption, informing future models and guiding researchers in prioritizing key factors for optimizing BEP.

The findings have three practical implications for real-world applications. First, the energy efficiency recommendations can be directly implemented by architects, engineers, and builders, facilitating the design of residential buildings optimized for hot climates and leading to reductions in energy consumption and operational costs. Second, the study’s data-driven approach supports informed decision-making by utilizing advanced simulation tools and predictive models, allowing stakeholders to evaluate different design scenarios and make choices that enhance energy efficiency. Third, the research offers clear guidance on the most effective BEs and materials, helping resolve discrepancies in the existing literature and providing practitioners with a clear understanding of strategies that contribute to energy savings.

The study has important implications for policy development. First, the findings can inform updates to building codes and standards, particularly in hot climates, by providing evidence-based recommendations for energy-efficient materials and design practices. Policymakers can use these insights to mandate adopting these practices, promoting widespread improvements in BEP. Second, the findings support the development of incentives, such as tax breaks or subsidies, to encourage the adoption of recommended energy-efficient practices. This can drive a broader implementation of sustainable building practices. Third, the success of BIM and ML highlights the need for policies supporting the adoption of these technologies in the construction industry, potentially including funding for technology adoption, training programs, and research initiatives.

However, despite these opportunities, a significant challenge remains in integrating advanced energy optimization strategies, particularly those involving BIM and ML, into existing building codes and practices. In many regions, especially in developing countries, current building codes may not yet support these technologies, complicating efforts to standardize their adoption across the industry.

5.4. Limitations of the Study

This study employed various tools and techniques to examine the designed energy efficiency methodology, focusing on applying machine learning algorithms. While this approach provides valuable insights, it also comes with inherent constraints in implementing tools and the assumptions made, which should be considered when interpreting the findings. First, the study is tailored to hot climates, with a particular focus on residential villas, and may not be directly applicable to other building types, such as high-rise buildings or commercial structures, or to regions with different climates. The findings, therefore, should not be generalized without further replication and examination.

Second, the study primarily relies on simulations with limited real-world validation, which introduces some degree of uncertainty in the results. The effectiveness of the machine learning models used is also influenced by the quality and size of the training dataset, which could limit the conclusions’ robustness. Additionally, the study does not extensively explore the cost implications of the recommended optimizations, which could affect their feasibility in practical applications.

Moreover, due to the limitations of the experiments, this study does not claim to assess or suggest that any particular building or city within the selected regions exhibits superior or inferior energy efficiency. The need for accurate temperature data from all stations across multiple seasons of the year further constrains the generalization of findings. Some cities in the regions studied exhibited higher average temperatures throughout the year, with summer temperatures peaking around 40 °C and winter temperatures remaining above 10°C. These variations and other potential factors, such as weather patterns and behavioral considerations, indicate that the results are context-specific.

5.5. Future Research Directions

Future studies should focus on the following areas to build upon this research. First, they should examine the applicability of the findings in different climatic regions, including cold and temperate climates, to generalize the results. Second, investigating energy optimization potential for various building types beyond residential villas, such as high-rise buildings, would provide a more comprehensive understanding. Third, implementing and monitoring the recommended optimizations in real-world settings would validate the simulation results and assess their practical effectiveness. Fourth, conducting detailed cost-benefit analyses would evaluate the financial feasibility of implementing energy-efficient practices. Fifth, studying the impact of occupant behavior on energy consumption would enhance the accuracy of energy performance predictions.

Additionally, future studies should explore how energy efficiency may vary under different climatic conditions, such as Dubai’s consistently warm and humid climate versus the larger diurnal temperature range and seasonal variation in Riyadh. Investigating the impact of innovative materials used in the residential sector on building energy consumption would also be a valuable avenue for further research. These considerations will help to contextualize further and expand the applicability of the findings in this study.

6. Conclusions

This research undertook a detailed examination of energy optimization analysis for residential buildings in the hot climates of Dubai and Riyadh by integrating BIM and ML. The study provides valuable insights into enhancing energy performance in these regions by constructing detailed BIM models and analyzing various building scenarios using energy simulations and predictive analytics.

The evaluation of different ML models, including GBM, RF, SVM, and LR, revealed that GBM consistently outperformed the other models. With the lowest error rates and highest R² values, GBM demonstrated superior accuracy and robustness in predicting EUI. The expanded dataset size further improved model performance, which indicates the importance of comprehensive data for reliable ML applications in energy performance analysis. These findings demonstrate the critical role of model selection and data volume in achieving accurate and reliable energy predictions.

The FIA using the GBM model identified roofs, exterior walls, and windows as the most influential factors affecting energy consumption. In RS, these BEs accounted for over 78% of the energy consumption, while in DS, they represented approximately 62%. The higher importance of RS indicates the necessity for superior thermal insulation and effective construction materials to manage extreme climatic conditions. Additionally, building features such as windows, plug load efficiency, and PV roofs were significant, emphasizing the need for a holistic approach to building design incorporating passive and active energy-saving measures.

The optimization analysis provided detailed insights into the best and worst building features regarding energy performance. Advanced insulation, such as VIP-insulated walls and reflective coatings, significantly reduced energy consumption. The use of green roofs and green cavity walls also proved effective in managing the cooling loads in these hot climates. Moreover, the evaluation of various BEs revealed that customized energy efficiency strategies that address specific climatic conditions and material selection are paramount in optimizing energy efficiency.

In conclusion, this study highlights the need for customized energy efficiency strategies to address Dubai and Riyadh’s unique climatic conditions and architectural practices. Integrating BIM and ML has proven to be a powerful approach to identifying and optimizing critical building features. These results offer valuable guidance for architects, engineers, and policymakers aiming to develop more energy-efficient buildings.