Artificial Intelligence-Based Decision Support System for Sustainable Urban Mobility

Abstract

As urban populations rise globally, cities face increasing challenges in managing urban mobility. This paper addresses the question of identifying which modifications to introduce regarding city mobility by evaluating potential solutions using city-specific, subjective multi-objective criteria. The innovative AI-based recommendation engine assists city planners and policymakers in prioritizing key urban mobility aspects for effective policy proposals. By leveraging multi-criteria decision analysis (MCDA) and ±1/2 analysis, this engine provides a structured approach to systematically and simultaneously navigate the complexities of urban mobility planning. The proposed approach aims to provide an open-source interoperable prototype for all smart cities to utilize such recommendation systems routinely, fostering efficient, sustainable, and forward-thinking urban mobility strategies. Case studies from four European cities—Helsinki (tunnel traffic), Amsterdam (bicycle traffic for a new city quarter), Messina (adding another bus line), and Bilbao (optimal timing for closing the city center)—highlight the engine’s transformative potential in shaping urban mobility policies. Ultimately, this contributes to more livable and resilient urban environments, based on advanced urban mobility management.

1. Introduction

Urban areas face numerous challenges, such as rapid population growth, the advent of new transportation modes, and escalating urban traffic, all contributing to heightened pollution and congestion. While innovative advancements in electronic and information technologies, leading to new mobility options like electric scooters, offer the potential for greener transportation solutions, their unregulated proliferation can exacerbate existing urban challenges. Consequently, there is an urgent need for informed, forward-thinking policies to address these challenges effectively.

As urbanization continues to deal with complex issues, the concept of smart cities emerges as a promising solution, improving life in information society [1]. Artificial intelligence (AI) [2,3] is central to this transformation, with the potential to revolutionize various urban life aspects, from optimizing transportation systems to managing environmental resources, enhancing governance, improving quality of life, fostering economic growth, and empowering citizens. Recommendation engines and decision support systems, underpinned by robust algorithms and extensive urban data, provide actionable insights, enabling the development of efficient, sustainable, and adaptable urban mobility frameworks.

Smart cities employ advanced electronics and information technologies, data analytics, and digital infrastructure to enhance the efficiency, sustainability, and quality of life for residents [4,5,6,7]. These cities integrate various smart solutions across multiple sectors, including transportation, energy, healthcare, governance, and education. Mobility strategies are one of the most relevant issues due to traffic overload in cities [8,9,10].

Cities offer various public transport options, including buses, trams, and metros, supplemented by private vehicles. However, traffic congestion remains a significant challenge, particularly in cities with high car usage and suboptimal road networks. While a complete infrastructure overhaul is impractical, intelligent planning for future demand is essential. Cities can perform analyses and test scenarios to understand the potential impacts of changes, such as closing specific streets and using computer simulations to mimic real-world traffic flow [11,12]. These simulations enable city institutions and stakeholders to make informed decisions about implementing changes to meet future traffic demands.

This study presents a policy evaluation tool, i.e., a system named DSSU (Decision Support System Urbanite). According to [13], a decision support system (DSS) is a computer-based system that combines data and decision logic as a tool for assisting a human decision maker. A DSS does not make decisions, but instead assists the human decision maker by analyzing data and presenting processed information. To this end, DSSU is designed to aid city institutions and stakeholders in making more informed decisions regarding traffic management. A practical application of this tool would be to assess the impact of policies that restrict private motorized vehicle usage during specific times of the day in city centers. These policies aim to encourage sustainable travel options such as public transport, walking, and cycling, thereby reducing dependence on private vehicles. The decision support model utilized in this study leverages computer simulations to determine the optimal street closure times for each district in Bilbao, Spain. The objectives include enhancing safety, improving public spaces, promoting active transportation, driving urban revitalization, achieving environmental benefits, and reducing noise pollution. Furthermore, the simulation results can be used as input to machine learning (ML) modules to speed up the simulations towards real-time testing of various city policies [14].

In addition to Bilbao, our prototype model is applied to three other EU cities as part of the H2020 URBANITE project (https://urbanite-project.eu/) (accessed on 12 September 2024): Helsinki, Amsterdam, and Messina. Each case study involves a unique urban mobility challenge, demonstrating the DSSU system’s versatility. In Helsinki, the DSSU system evaluates the potential impact of constructing a new tunnel on nearby urban traffic, assessing outcomes such as pollution reduction, noise decrease, and traffic congestion alleviation. In Amsterdam, the focus is on a new city quarter and its implications for bicycle traffic in the city center, determining if the addition would impede the current flow of bicycle traffic. In Messina, the system analyzes the introduction of a new bus line to see if it would enhance overall traffic flow proportionally.

DSSU has been designed to provide a comprehensive analysis of all four cities, including objective metrics and subjective preferences specific to the city teams’ goals. The key questions addressed by the city teams are:

• Bilbao: Should private traffic be restricted in the city center; and if so, when, to achieve the most significant positive impact?
• Helsinki: How will a new tunnel affect urban traffic in terms of pollution, noise, and congestion?
• Amsterdam: Will introducing a new city quarter result in stagnant bicycle traffic in the city center?
• Messina: Will adding a new bus line improve overall traffic flow effectively enough to support the introduction?

By successfully and transparently addressing these questions, the policy evaluation by DSSU is expected to offer a robust framework for city planners to make data-driven and context-sensitive decisions to improve urban mobility and sustainability. This is the central motivation of the study presented.

This study is closely aligned with the integration of electronic systems within the broader framework of smart cities. Our research contributes to this field by introducing an AI-based simulation tool for urban mobility, which fundamentally depends on the robust infrastructure of electronic systems, IoT devices, networks, and sensors—key components of smart city architecture. These electronic systems provide the essential foundation for collecting, processing, and analyzing vast amounts of data, which our system leverages to optimize urban mobility.

Although our primary focus is on the software and AI-driven aspects of DSSU, its success, and effectiveness are deeply intertwined with the performance and integration of the underlying electronic hardware. The intelligent management and real-time optimization capabilities of DSSU are entirely reliant on the seamless operation of the embedded electronics, sensors, and communication networks that support smart city infrastructure.

The strong dependence of DSSU on electronic systems highlights our work’s role in improving the integration and efficiency of these technologies in smart cities. This study directly contributes to advancing fields such as computer science and engineering, networks, artificial intelligence, and smart cities.

To the best of our knowledge, no existing scientific publication describes a DSS that facilitates similar subjective, AI-based evaluation and comparison of current urban mobility scenarios with proposed modifications. Additionally, there is no open-source solution available that offers the functionality presented in this paper.

From a scientific perspective, this study hypothesizes that an AI-based decision support system can be effectively algorithmically designed and implemented as a functional software system and significantly enhance city planners’ ability to address urban mobility challenges. Specifically, it is hypothesized that AI-driven modeling and multi-criteria scenario evaluation will lead to more informed and effective policy decisions, improving traffic flow, reducing emissions, and enhancing the overall quality of urban life. By testing this hypothesis through the deployment of DSSU in various European cities, this research seeks to provide valuable insights into the potential of AI-driven tools in urban mobility planning.

The structure of the paper consists of the Introduction, followed by related work, and Section 3, Materials and Methods, where the methods and data are presented. Section 4 describes the results while Section 5 concludes with a discussion.

2. Related Work

One of the most representative overviews of smart city literature can be found in [15], where a search in the Scopus database using paper keywords was performed based on three queries with the following hits for the years 2021, 2022, and 2023: “smart city” (4282, 3537, 3306), “artificial intelligence” (33,735, 35,689, 43,409), and both (282, 230, 246). These numbers indicate that the “smart city” keyword declined significantly in the last three years while “artificial intelligence” increased by one-third. AI is likely the most relevant field of research and development in recent years, with the EU facing stiff competition from China and the USA [16].

Finally, 157 articles were fully reviewed and summarized from 2021 till 2023, including several months in 2024, divided into the following categories: smart governance, smart economy, smart mobility, smart environment, smart living, and smart people [17,18,19].

A Search for 2023 and 2024 till mid-June 2024 revealed the 102 most relevant hits when looking for smart cities. None of the papers revealed a major similarity with our study. However, several papers bear some similarities to this paper. For instance, Ref. [20] deals with the multi-criteria Decision Making Trial and Evaluation Laboratory (DEMATEL) method used to identify the interdependent relations between smart cities’ thematic areas, structuring a diagram of cause-and-effect relations using threshold quantification values. For the degree of influence, the most influential thematic areas are coexistence and reciprocity, living environment and infrastructure, entrepreneurship, and healthcare. Also, the cause-and-effect analysis identified governance and engagement, education and training, and mobility as central thematic areas for smart city management.

The study in [21] focuses on applying multi-criteria decision-making (MCDM) methods to select transport means in metropolitan areas, emphasizing sharing mobility. It aligns with our work by addressing the complexities and subjective aspects of decision making in urban mobility contexts using multiple criteria. Other studies employ MCDM to address smart mobility topics, such as evaluation [22,23], promotion [24,25] and planning [26] of smart mobility projects, assessment of sustainable mobility in cities [27,28,29], and assessment of mobility of employees [30,31]. Overall, these studies indicate a widespread use (and usefulness) of MCDM methods in the area of sustainable urban mobility. Such studies generally include economic, environmental, and social criteria. However, the selection of individual criteria is problem-dependent and varies substantially among studies. A wide variety of MCDM methods is used too, including WSM, AHP, ANP, MAUT/MAVT, TOPSIS, MACBETH, PAPRIKA, BWM, VIKOR, PROMETHEE, ELECTRE, UTA, DRSA, and DEX (see [32,33,34] for more information about these methods).

Multi-objective decision making is also a major theme in [35]. The paper evaluates transportation projects and mobility solutions based on multiple criteria to aid decision-making processes in urban environments.

In [36], the use of big data analytics and context-aware computing is explored to enhance sustainability in smart cities. It discusses multi-criteria decision-making frameworks in integrating advanced technologies to support sustainable development goals. The research highlights subjective decision-making aspects by considering diverse criteria and stakeholder preferences.

3. Materials and Methods

The DSSU decision support system developed in this research is designed to evaluate and enhance mobility policies within urban environments, utilizing a robust framework that combines traffic simulations and multi-criteria decision models. This methodology integrates several essential components: data preparation, traffic simulation, metrics calculation, data discretization, and policy evaluation using the decision modeling system called DEXi. The primary objective of DSSU is to equip urban planners with the necessary tools to make well-informed decisions based on detailed traffic scenario simulations and their impacts on key performance indicators (KPIs).

Data preparation involves collecting and processing various datasets such as traffic flow, vehicle types, road network configurations, and socio-demographic information about the urban population. This step ensures that the input data are accurate, comprehensive, and suitable for realistic simulation scenarios.

Following data preparation, detailed traffic simulations were conducted using agent-based models, which simulate the daily travel behaviors and interactions of individual agents (city inhabitants). These simulations accounted for a variety of factors, including travel demand, route choices, and transportation modes, providing a dynamic and granular view of traffic patterns across different policy scenarios.

The baseline scenarios were meticulously developed using publicly available data that accurately reflect the current conditions in the cities, supplemented by data obtained from city teams. Most of the data were sourced from city teams and existing databases, ensuring a robust foundation for the simulations. Detailed city maps were acquired via OpenStreetMap, and for modeling travel plans, cities either provided their data or, where necessary, a synthetic population was generated using data from the European Union Statistics on Income and Living Conditions (EUSILC). This synthetic population was further refined with marginal distributions specific to the pilot cities. An example of the synthetic data is shown in

Table 1. An example of the results of synthetic data generated from the travel demand model.

Person Group	Age	Activities	Transport Mode
Young students	<16	Home, education	Public transport, bicycle, walk
Young students	≥16, <18	Home, education	Car, public transport, bicycle, walk
University students	≥18, <24	Home, education, work, leisure, shopping, other	Car, public transport, bicycle, walk
Regular workers	≥24, <65	Home, work, leisure, shopping, other	Car, public transport, bicycle, walk
Delivery service workers	≥24, <65	Home, work	Bicycle
Elderly	≥65	Home, leisure, shopping, other	Car, public transport, bicycle, walk

. Additionally, whenever available, data from the pilot cities were incorporated to enhance the realism and accuracy of the models. Our experience shows that large cities often manage vast amounts of data, making this integration both feasible and beneficial. In summary, the data used in this study were a combination of public sources and city-specific data, which may or may not be publicly accessible. Importantly, our source code is available as open source.

However, the proposed solutions were not tested in real-world settings due to several practical constraints. For instance, testing traffic modifications in anticipation of a forthcoming tunnel would be both impractical and potentially misleading. At the same time, it is important to note that nearly all city-specific data used in the simulations were derived from actual inputs gathered through sensors, electronic devices, networks, IoT systems, and other data sources. For example, agent plans were based on real-world observations, such as a person visiting a school on a typical workday. From these data, a software agent representing a “scholar” was created, and it was programmed to attend school at specific times. If the usual route for this agent was disrupted or improved, the agent would attempt to find an alternative best route. Thus, while the system operates through simulations and decision-making processes, it is firmly rooted in actual observed behaviors and performance data detected through advanced sensors and electronic systems.

In practical terms, the software team meticulously sourced data from city teams, publicly available datasets, and specialized databases, focusing on specific time specifications to build a foundation for one round of simulations. The timing of data collection was carefully planned to maximize both the accuracy and relevance of the analysis. Data were collected at multiple intervals throughout the three-year project, ensuring that each dataset captured the most current urban conditions to support extended simulations, decision support, and comprehensive analysis. Recognizing the inherent variability in traffic patterns, the team strategically timed data collection to coincide with periods that represented typical traffic conditions, such as those observed on regular working days. This approach ensured that the simulations were not only grounded in real-world scenarios but also temporally aligned with actual urban dynamics. As a result, the insights derived were highly applicable to both existing conditions and future modifications, offering practical guidance for urban planning and decision making, rather than theoretical solutions disconnected from reality.

The inclusion of demographic data such as age, gender, income levels, typical goals and motivations, and employment status is crucial in accurately modeling travel demand within the simulation framework. Age and employment status are particularly significant, as they directly influence the types of activities assigned to travel agents, such as school attendance for younger individuals and work-related travel for employed adults. These activities are inherently tied to specific locations, which in turn determine the origins and destinations of trips within the urban environment. For example, school-age individuals are mapped to school-related travel during school hours, while employed adults are mapped to work-related travel during typical business hours. This assignment not only reflects realistic daily routines but also ensures that the simulation accurately captures the spatial and temporal distribution of travel demand in case of route modifications.

Furthermore, there is a correlation between these demographic factors and the chosen KPIs. For instance, the number of car and bicycle trips, average trip durations, and congestion levels are all influenced by the distribution of age groups and employment statuses across the population. Younger populations might contribute to higher school bus usage or cycling while working adults might contribute more significantly to rush hour traffic and longer commutes. Additionally, income levels can affect mode of transportation choices, with higher-income individuals potentially opting for private vehicle use, which in turn impacts emissions and congestion KPIs. By incorporating these demographic variables, the simulation is better equipped to reflect the complexity of urban mobility patterns and to predict the impacts of proposed policies on different segments of the population.

Metric calculation is another critical component of the DSSU system, where numerous KPIs are derived from the simulation outputs. These KPIs include, but are not limited to, CO₂ emissions, NOx emissions, particulate matter (PM) emissions, the number of car and bicycle trips, average trip durations, average bus speeds, and congestion levels. Collectively, these metrics provide a comprehensive assessment of the environmental, operational, and social impacts of the proposed mobility policies.

Data discretization is employed to convert continuous KPI values into discrete categories, facilitating integration into the decision model. This step involves relativizing and mapping the KPI values into predefined ranges, enabling a more straightforward comparison and evaluation process.

The core of DSSU is the policy evaluation phase using DEX, a qualitative multi-criteria decision modeling method [37]. DEX is supported by the software DEXi [38] (https://dex.ijs.si/), in which multiple criteria are structured in a hierarchical model, allowing for a detailed subjective analysis of each policy scenario. By comparing simulated scenarios against a baseline, DSSU identifies the most effective policies and highlights areas needing improvement.

The iterative evaluation process, supplemented by the ±1/2 analysis method, further refines policy assessments by pinpointing specific KPIs that require enhancement to achieve optimal outcomes. This detailed and systematic approach ensures that urban planners can develop and implement mobility policies that not only improve urban mobility and sustainability but also align with broader strategic goals for urban development.

As presented in Figure 1, the methodology consists of simulations, KPI calculation, data preparation, and policy evaluation.

3.1. Simulations Data

Data preparation for the simulations involves several steps to ensure that the input for the MATSim simulator [39] (https://www.matsim.org/ accessed on 12 September 2024) is accurate and comprehensive. The necessary files include the city network file, vehicles file, transit vehicles file, agents plans file, and the transit schedule file.

The city network file is created using OpenStreetMap (https://www.openstreetmap.org/ accessed on 12 September 2024), which provides detailed geographical and infrastructural data, including road networks, intersections, and traffic regulations. These data are further processed to represent the physical layout of the city accurately, ensuring that the simulation can account for real-world constraints and opportunities.

For the vehicles file, data on the types and numbers of vehicles operating within the city are collected and processed. This includes private cars, buses, bicycles, and other modes of transport. The transit vehicles file, specifically, details the public transport options available, including buses and trams, providing critical information on their capacities, speeds, and operating schedules. These data were collected for the purposes of the project during the years 2020 to 2022.

The agents’ plan file is generated by modeling the daily activities and travel behaviors of individual agents, i.e., city inhabitants. This involves synthesizing data from surveys and marginal distributions, processed using iterative proportional fitting (IPF) to achieve a realistic demographic distribution. Each agent’s plan includes details such as departure times, destinations, and chosen modes of transport, reflecting the diverse travel patterns within the city. Data used to generate the agents’ plans were gathered from the EUROSTAT SILC 2013 dataset for the relevant countries, and the marginal distributions for each city were gathered in the years 2019 to 2021.

The transit schedule file, based on the General Transit Feed Specification (GTFS) [40], provides comprehensive information on public transport timetables, routes, and stops. These data ensure that the simulation accurately represents public transport availability and scheduling, which is crucial for evaluating policies that impact transit operations. These were provided by the city authorities for the project in 2022.

By integrating these files, the data preparation process ensures that the MATSim simulator can operate with a high degree of accuracy and relevance. This allows for the detailed analysis of various traffic scenarios, enabling urban planners to make informed decisions aimed at improving urban mobility and sustainability.

3.2. Travel Demand Generation

Travel demand generation is a crucial step in modeling urban mobility, involving the creation of synthetic data representing the daily activities of city inhabitants, referred to as agents. This process starts with collecting and integrating comprehensive survey data, including various socio-demographic attributes such as age, gender, income levels, and employment status [10]. These data points are crucial for accurately reflecting the diverse population and their travel behaviors within the urban area. Often, they are gathered from telecom operators since most inhabitants possess mobile phones. Based on their movements and endpoints, several relevant roles can be deduced; e.g., inhabitants entering educational institutions and staying there for a longer time; visitors of shopping malls, etc.

To ensure that the synthetic population accurately mirrors the real-world demographic distribution, IPF is employed. It adjusts the synthetic data to match marginal distributions obtained from census or survey data, ensuring that the generated population is representative of the actual demographic structure. This method not only enhances the realism of the simulation but also improves the reliability of the results by providing a solid foundation for subsequent modeling steps.

The core of travel demand generation lies in the creation of origin–destination (O-D) matrices. These matrices capture the travel patterns of agents by detailing where trips originate and terminate, effectively mapping out the flow of people across the urban landscape. O-D matrices are derived from travel diaries or survey data, which record individual trips, including purpose, start and end locations, and travel modes. This information is crucial for understanding peak travel times, common routes, and the overall demand for different modes of transportation.

In addition to O-D matrices, individual attributes play a significant role in shaping travel behavior. Attributes such as household size, car ownership, employment status, and school locations influence daily travel plans [41]. By incorporating these factors, the model can simulate a wide range of travel behaviors, from commuting to work or school to leisure trips and shopping.

Advanced modeling techniques are used to generate daily travel plans for each agent. These techniques include activity-based modeling (ABM), which simulates the sequence of activities and trips an individual undertakes throughout the day. ABM considers the constraints and preferences of agents, such as work hours, school schedules, and preferred travel modes, providing a detailed and dynamic representation of urban mobility.

Furthermore, the integration of real-time data, such as traffic conditions and public transport schedules, can enhance the accuracy of travel demand models. Real-time data help in adjusting travel plans dynamically, reflecting changes in travel conditions and enabling the simulation of scenarios like traffic congestion or public transport disruptions.

Overall, travel demand generation is a comprehensive process that combines demographic data, travel behavior analysis, and advanced modeling techniques to create a realistic and dynamic representation of urban travel patterns. This foundational step is essential for developing effective urban mobility solutions and policies. For this study, the data from cities were studied and enriched whenever needed. One of the emphases was on the interoperability of the obtained data, meaning all data were transformed into a compatible form although describing different types of movement and observed tasks.

3.3. Traffic Simulation Process

In planning mobility modifications, city decision makers resort to various solutions, e.g., applying experience from other solutions, polling residents, and executing software simulations. There are numerous city mobility simulation tools, including open-source packages. Some are listed here and an overview is provided in

Table 2. Comparison of traffic simulation software. For each simulation tool mentioned we list whether it is open source (required for extensibility), supports multimodal simulations (required to analyze cyclists, public transport, and car traffic), supports microscopic simulations (required for detailed analysis). Required features are supported by most modern traffic simulation tools and the final decision for MATSim was based on ease of extension and integration.

Simulation Tool	Open Source	Multimodal	Microscopic	Comment
MATSim 12.0	Yes	Yes	Yes	Easy to extend and integrate
SUMO 1.12	Yes	Yes	Yes	Hard to extend and integrate
TRANSIMS 7.4	Yes	Partially	Yes	Development inactive
Anylogic 8.7	No	Yes	Yes	Commercial
PTV Vissim 2022	No	Yes	Yes	Commercial

:

• MATSim (Multi-Agent Transport Simulation): A powerful agent-based transport simulation framework. It models individual travelers as agents with specific preferences and behaviors, allowing for detailed analysis of transportation systems. Researchers and practitioners can use MATSim to simulate various scenarios, test policies, and optimize urban mobility [42].
• Simulation of Urban Mobility (SUMO): An open-source, highly portable, microscopic road traffic simulation package designed to handle large road networks. It is a free alternative to MATSim [43].
• TRANSIMS (Transportation Analysis and Simulation System): An integrated set of tools developed for regional transportation system analyses. It is a free and open-source alternative to MATSim [44].
• Anylogic: A cross-platform multimethod simulation modeling framework that supports agent-based, discrete-event and system dynamics simulation methodologies. It is a popular commercial alternative to MATSim [45].
• PTV Vissim: aA proprietary traffic simulation software solution, commonly used in urban and traffic planning. Supports microsimulation with advanced features such as simulating pedestrians in 3D spaces and modern 3D visualizations [46].

Simulation frameworks can show the results of simulations in a city before and after the intended modification; however, there are several issues:

• There can be a large number of potential policies, e.g., regarding the start and end of a closure of a particular street;
• In addition to standard attributes such as traffic congestion, each city prefers its own preference criteria, i.e., a subjective evaluation function.

Given these assumptions, and the desire to design an open-source interoperable smart city system, we chose MATSim and upgraded it with several modules, including a novel smart city recommendation module, enabling subjective comparisons and analysis of potential mobility strategies. Among the needed modifications were several databases, e.g., the handbook of emission factors for road transport (HBEFA), that enabled computing consumption of vehicles, including the cold starts.

The MATSim simulator, an agent-based simulation framework, is employed to model the daily activities and travel plans of agents in the four EU cities. The simulation iterates through daily plans, optimizing routes and travel modes to reduce congestion and improve traffic flow [6]. However, the internal scenarios for the four cities differ. For instance, in Bilbao, various street closure scenarios are simulated by altering access to certain roads and analyzing the resulting impacts on traffic dynamics and emissions.

Several types of traffic simulations are used, including microscopic (detailed individual vehicle interactions), macroscopic (overall traffic flow), mesoscopic (group behaviors of vehicles), and agent-based simulations (detailed modeling of all transportation system participants). The agent-based approach is particularly suitable for this study due to its detailed representation of individual travel behaviors and interactions [47]. The results of the simulations are presented using various graphical packages; e.g., Figure 2 is generated by the Simunto Via 22.1 (https://docs.simunto.com/via/) [accessed on 12 September 2024] graphics package.

3.4. KPI Metric Calculation

Metric calculation involves deriving KPIs from the simulation output to evaluate the performance of different mobility policies. The KPIs include CO₂ emissions, NOx emissions, particulate matter (PM) emissions, number of car trips, number of bicycle trips, average trip duration, average bus speed, number of “stuck and abort” events, and number of congested links [11]. These metrics provide a comprehensive view of the environmental and operational impacts of the simulated policies.

These metrics offer valuable insights into how different mobility strategies impact urban areas. For instance, reductions in CO₂ and NOx emissions indicate improved air quality, while an increase in the number of bicycle trips suggests a shift towards more sustainable transportation modes. Similarly, metrics such as average trip duration and bus speed help assess the efficiency of the transportation network.

By analyzing these KPIs, urban planners can identify the effectiveness of different mobility strategies in reducing pollution, improving traffic flow, and enhancing overall urban mobility. This thorough evaluation supports the development of informed, sustainable, and efficient urban transportation policies, ultimately contributing to the creation of smarter, more livable cities.

3.5. Data Discretization

The continuous KPI values from the simulations are discretized to fit the decision model’s requirements, a critical step for effectively utilizing the DEXi decision modeling system in the next stage. This process involves two primary steps: relativization and discretization. Relativization calculates the relative difference between the KPIs of two simulations, providing a comparative basis for analysis. Discretization then maps these relative differences into predefined ranges, transforming continuous data into discrete values suitable for the decision model.

Relativization is particularly important as it normalizes the KPIs, allowing for a fair comparison between different policy scenarios. This step ensures that the differences in performance metrics are accurately represented, facilitating a more nuanced evaluation of each policy’s impact. Discretization further refines these data, converting the continuous KPI values into a format that the DEXi model can process. This step involves categorizing the relative differences into specific ranges, simplifying the data, and making them more manageable for analysis.

For instance, in the context of urban mobility, continuous KPIs such as CO₂ emissions or average trip durations are transformed into discrete categories like low, medium, and high. This categorization aids in highlighting significant changes and trends that might not be as evident in raw, continuous data. By structuring the data this way, urban planners can more easily interpret the results and make informed decisions about which policies to implement.

3.6. Decision Model Creation

The decision model is created using the qualitative multi-criteria decision-modeling method DEX [37]. In DEX models, multiple criteria are represented by discrete variables called attributes. Attributes are structured in a hierarchy, where terminal nodes represent inputs and roots represent outputs of the decision evaluation process. Transitions from input to output attributes are modeled in terms of decision rules. The DEXi software [38] represents a tool for multi-criteria decision analysis (MCDA). It facilitates the evaluation of complex decision-making scenarios by breaking down the problem into manageable criteria and sub-criteria and formulating decision rules in terms of decision tables. This hierarchical structure allows for a systematic assessment of each criterion’s impact on the overall decision-making process.

In the context of urban mobility, the model evaluates different policy scenarios based on discretized KPIs, producing a comprehensive assessment of policy quality. Each KPI, such as CO₂ emissions, average trip duration, and number of bicycle trips, is first discretized into predefined ranges to simplify the comparison between different scenarios. The DEXi software then uses these discretized values to assess the relative performance of each policy option.

The DEX method involves constructing a decision model that integrates various criteria relevant to urban mobility. The hierarchical tree structure in DEXi is particularly advantageous as it allows decision makers to visualize and understand the relationships between different criteria and their sub-criteria. This structure includes decision tables that assign scores to each criterion based on its performance, ultimately aggregating these scores to provide an overall assessment of each policy scenario.

For example, a policy scenario aimed at reducing traffic congestion might include criteria such as emission reductions, improvements in average bus speeds, and increases in bicycle trips. Each of these criteria is evaluated separately, and their scores are aggregated to determine the overall effectiveness of the policy. The hierarchical tree structure ensures that each aspect of the policy is considered, providing a balanced and comprehensive evaluation. Figure 3 represents an example tree of model attributes. The selection and structuring of attributes depends on the task at hand and may be different for different urban mobility projects.

The DEXi model’s ability to handle both discretized quantitative and qualitative data makes it a powerful tool for urban planners. By incorporating expert knowledge and stakeholder preferences into the decision-making process, DEXi enhances the robustness and reliability of the policy evaluations [48]. This approach ensures that the selected policies not only meet technical performance standards but also align with broader sustainability and livability goals.

In summary, the DEX method, supported by the DEXi software, provides a structured and systematic approach to evaluating urban mobility policies, not only in objective but also in subjective ways. By integrating multiple criteria and using a hierarchical tree structure, the decision model offers a comprehensive assessment of policy scenarios, aiding urban planners in making informed and balanced decisions for sustainable urban development.

3.7. Evaluation Process

Policy evaluation is a critical phase in the decision-making process, aimed at determining the effectiveness of various urban mobility policies. This process involves grading each simulation scenario against a baseline using the software called DEXiEval 5.3 (https://dex.ijs.si/dexisuite/dexieval.html) [accessed on 12 September 2024]. DEXiEval is a command-line tool that employs a DEX model to systematically evaluate the performance of each scenario by comparing its KPIs to those of the baseline scenario. These KPIs might include metrics such as CO₂ emissions, average trip duration, and the number of bicycle trips, among others.

The package ranks the scenarios based on their performance, identifying both the best- and worst-performing policies. This ranking is achieved through an iterative comparison process, where each scenario is compared with others. Scenarios that outperform the baseline are incremented in a counter, thus highlighting those policies that offer superior outcomes.

The iterative comparison process involves multiple rounds of evaluation, where scenarios are continuously compared to refine the understanding of their relative performance. This method ensures that the most effective policies are identified through a thorough and detailed analysis. By incrementing a counter for each scenario that outperforms the baseline [49], DEXiEval provides a clear and quantifiable measure of policy effectiveness. Figure 4 shows an example of policy evaluation in DEXi, where the changes are also evaluated through simple wording such as “same” or “better”.

Moreover, this evaluation process is not just about identifying the top-performing scenarios but also about understanding the reasons behind their success. This helps urban planners and policymakers to pinpoint which aspects of a policy contribute most to its effectiveness. By leveraging this package, urban planners can make well-informed decisions that enhance urban mobility and contribute to sustainable city development. This structured approach to policy evaluation ensures that the chosen policies are not only effective but also aligned with the broader goals of urban sustainability and quality of life improvement.

3.8. ±1/2 Analysis

When comparing modifications with the baseline, they are often of the same qualitative class. To further differentiate between them, the so-called ±1/2 analysis method is employed by identifying KPIs that need improvement to match the best-performing scenario. This involves incrementing or decrementing KPI values in small steps to see their effect on the overall policy quality, providing detailed insights into which aspects of a policy need adjustment [4].

Therefore, the purpose of this incremental adjustment is to identify which KPI changes lead to significant improvements in policy performance. For instance, if a particular scenario shows that reducing CO₂ emissions slightly can greatly enhance the policy’s effectiveness, then this becomes a focal point for policymakers. Conversely, if increasing the number of bicycle trips by a small margin does not substantially affect the overall policy quality, then this aspect might be deprioritized. At the same time, one should be aware that the subjective space of preferences need not be proportional, i.e., the same modification in one space might cause unproportional modification in the other. The “1/2” in the name of the method refers to making a one- or two-step change in an individual criterion, considering its discrete value scale. Such steps correspond to negative or positive changes in the corresponding KPIs, expressed as percentages of the baseline value. Exact percentages depend on individual KPIs but are typically in the range of 2% to 20%.

This method allows for a granular understanding of policy dynamics, ensuring that each KPI’s influence is thoroughly examined. The iterative nature of ±1/2 analysis means that policymakers can test various combinations and magnitudes of KPI adjustments, providing a robust framework for continuous improvement. This approach is especially valuable in complex urban environments, where multiple factors interplay to determine the success of a policy.

By employing the ±1/2 analysis, urban planners and decision makers gain additional means to fine-tune their strategies, ensuring that the implemented policies are not only effective but also optimized for the specific needs and conditions of their cities. This method enhances the precision of policy evaluations, leading to more informed and targeted urban mobility solutions.

4. Results—Case Studies: Steering Mobility Policies in Four European Cities

4.1. Amsterdam

Amsterdam, renowned for its extensive bicycle network, faces challenges in maintaining and enhancing its bikeability amidst ongoing urban development. The city’s goal is to balance the introduction of new districts with the preservation and improvement in bicycle traffic flow, ensuring safety and efficiency for cyclists.

Problem description: The primary challenge for Amsterdam is integrating the new district while maintaining seamless bicycle traffic in the city center. The introduction of Amsterdam Noord, designed to accommodate approximately 70,000 new residents, raises concerns about potential disruptions to bicycle traffic, increased congestion, and compromised safety for cyclists. City planners need to assess whether the new district will hinder current bicycle mobility and identify strategies to mitigate any negative impacts.

The simulations revealed that the introduction of Amsterdam Noord has mixed impacts on bicycle traffic. While the new district provides opportunities for expanded cycling routes and infrastructure, it also poses risks of increased congestion at critical junctions. The analysis shows that without strategic interventions, the flow of bicycle traffic could be hindered (but not necessary), especially during peak hours. The analysis also provided recommendations highlighting effective strategies to mitigate these impacts, for instance, enhancing bike lanes, implementing dedicated cycling paths in the new district, and optimizing traffic signals that can help maintain smooth traffic flow.

The left side of Figure 5 illustrates a city map and the right side is a four-dimensional time-view of bicycle traffic for both the baseline and Amsterdam Noord scenarios.

Figure 6 displays a spider web chart that compares the current situation—already prone to bicycle congestion during rush hours—with the scenario after the introduction of the Amsterdam Noord district. The spider chart includes axes corresponding to individual criteria, highlighting differences between the baseline and the studied scenario. Values around the scale value 3 indicate no difference, while lower and higher values generally represent worsening and improvement, respectively.

Figure 5 and Figure 6 suggest that with appropriate measures the introduction of an additional district may not significantly exacerbate bicycle traffic problems.

The following are among the most relevant analysis results, as generated by DSSU and slightly tidied up by the authors.

According to the DEX multi-criteria model, the simulation scenario shows no significant degradation (when properly introduced) over the baseline one and is considered equal.

Specific recommendations:

• ±1 suggestion: No recommendation could be made using the ±1 analysis of the decision-making system.
• ±2 suggestion: To change the KPI of mobility policy quality by 10%, bikeability should be improved by 20%.
• To change the KPI of local by 10%, bikeability should be improved by 20%.
• To change the KPI of local bike infrastructure by 10%, bikeability should be improved by 20%.

The ±1/2 analysis once more indicates that a proper addition of the district should not cause major problems for bicycle mobility.

Conclusions: The case study of Amsterdam underscores the importance of strategic planning and the use of advanced decision support systems in urban mobility projects. The integration of Amsterdam Noord presents both challenges and opportunities for enhancing bicycle mobility. By leveraging comprehensive simulations and multi-criteria decision analysis, city planners can develop targeted interventions that preserve the city’s bikeability while accommodating new urban developments. The findings from this study serve as a valuable guide for other cities facing similar challenges, demonstrating that expanding cities need not decrease the quality of mobility, and therefore, quality of living, if another district is introduced properly.

4.2. Bilbao

Bilbao, a city known for its vibrant culture and economic activity, faces significant challenges in managing traffic congestion in its central areas. The primary focus of this case study is on Moyúa Square, a critical junction that experiences heavy traffic flow, leading to frequent congestion and increased pollution levels. This congestion not only leads to increased travel times and frustration for commuters but also contributes to higher levels of air pollution and noise. The goal is to explore strategies for optimizing traffic flow and improving the overall mobility experience in the city center.

Problem description: The challenge is to find a balance between maintaining accessibility and reducing traffic-related issues by closing the city center for specific periods, thereby enhancing the quality of life for residents and visitors.

The simulations revealed that partial or complete closure of Moyúa Square to private vehicles can influence congestion and improve traffic flow. Key findings include:

• Partial closure: Implementing a partial closure during peak hours can lead to a notable decrease in traffic congestion and travel times, in particular in the city center. This scenario also shows a reduction in CO₂ and NOx emissions, contributing to improved air quality. Around 10 specific closures from 1500 tested appear most interesting. It should also be noted that one figure represents only a couple of scenarios, i.e., the existing one and one or a few modifications, while many more scenarios were tested for each city.
• Complete closure: A complete closure of Moyúa Square to private vehicles, while more drastic, results in the most substantial reduction in congestion and emissions in the city center. However, this scenario requires robust alternative transportation options, such as enhanced public transport and bicycle infrastructure to accommodate displaced traffic. Side-effects of total closure might introduce new issues, significantly reducing the obtained benefits, e.g., more congestion in other parts of the city.
• The DSSU also provided specific recommendations for optimizing traffic signals and rerouting strategies to further enhance traffic flow and reduce delays.

Relevant results are presented in Figure 7. In this case, the spider web and the different scenarios are more unbalanced, indicating that there are important differences related to the parameters of the closures.

Conclusions: The findings highlight that strategic interventions, such as partial or complete closures of congested areas and enhancements in public transport, can lead to significant improvements in traffic flow and air quality. However, specific partial closures provide significantly better results than others on average, including complete closures. Various other simulations indicate that closure in part of the city in general results in more traffic in other areas.

4.3. Helsinki

Helsinki, a city known for its efficient public transport system and commitment to sustainability, faces specific challenges in managing traffic congestion around its West Harbour area. This area serves as a crucial transport hub, with heavy traffic from ferry arrivals and departures contributing to significant congestion and pollution. The West Harbour area in Helsinki experiences frequent congestion due to the high volume of ferry traffic, which leads to delays, increased pollution, and reduced mobility efficiency.

Problem description: The primary goal of this case study is to evaluate the potential impact of constructing a new tunnel to alleviate congestion and improve overall traffic flow in the area. The challenge lies in implementing the tunnel as an effective solution to manage congestion without disrupting the overall transport network, necessitating a thorough evaluation to ensure its feasibility and effectiveness.

The two scenarios, shown in Figure 8, indicate that the introduction of the tunnel might not provide a major overall improvement on its own unless accompanied by additional measures. The highly unbalanced spider web and similar scenarios in the right-hand figure indicate that the KPIs may change in both positive and negative ways, leading to an insignificant overall effect. Consequently, the tunnel introduction has to be accompanied by additional constructions if the quality of city life is to be improved.

The simulations indicated that constructing a new tunnel in the West Harbour area can significantly reduce congestion and improve traffic flow, but several KPIs tend to degrade. Key findings include:

• Tunnel construction: The new tunnel effectively influences congestion around the West Harbour by providing an alternative route for ferry traffic to access the highway. This scenario shows a notable decrease in CO₂ and NOx emissions in the surface streets due to smoother traffic flow and reduced idling times.
• Traffic and pollution redistribution: The tunnel helps redistribute traffic more evenly across the road network, reducing bottlenecks and improving travel times. Enhanced connectivity between the harbor and the highway leads to more efficient movement of goods and passengers. However, the analysis also revealed potential trade-offs and side effects, such as the need for additional infrastructure investments to support the tunnel air quality, extraction of pollutants, and the importance of integrating the tunnel with existing public transport options to maximize benefits. Overall, the amount of pollution in the city tends to stay the same unless additional solutions are introduced.

Conclusions: The Helsinki case study highlights the potential of tunnel construction to mitigate traffic congestion in critical urban areas. However, the simulation results in general did not meet the initial high expectations. While pollution is redirected from city streets into the tunnel, it becomes more concentrated, adversely affecting air quality for drivers. Further analyses are needed to explore solutions such as improving air quality within the tunnel and extracting the polluted air from the city.

4.4. Messina

Messina, a Mediterranean city characterized by its unique geographical constraints and bustling urban life, faces significant challenges in enhancing its public transport network. The primary focus of this case study is on the introduction of a new bus line aimed at improving accessibility and connectivity between different parts of the city. The goal is to evaluate the impact of this new bus line on overall traffic flow, public transport usage, and urban mobility.

Problem description: The introduction of a new bus line in Messina is intended to address these issues by providing a more efficient and accessible public transport option. However, the city needs to assess whether this new bus line will effectively improve overall traffic flow and public transport usage, and how it will impact existing transport patterns.

The two similar scenarios in Figure 9 indicate that the bus lane introduction has to be accompanied by additional constructions if the quality of city life is to be improved.

The simulations indicated that the introduction of the new bus line in Messina might significantly improve public transport connectivity and overall traffic flow, but only if accompanied by additional measures such as a campaign to use public transport instead of private vehicles. Key findings include:

• New bus line implementation: The new bus line provides better connectivity between previously underserved areas, leading to increased public transport usage and reduced reliance on private vehicles. This scenario shows a decrease in CO₂ and NOx emissions due to a shift from private vehicle use to public transport.
• Enhanced accessibility: However, the analysis also highlighted the need for complementary measures, such as improving the frequency and reliability of bus services, integrating the new bus line with other modes of transport to maximize its benefits, and implementing a campaign to reduce private transport usage.

Conclusion: The case study of Messina demonstrates the potential of strategic public transport enhancements to improve urban mobility and reduce environmental impacts. While it might be assumed that adding a bus line will automatically enhance the quality of life in the city, the results indicate a more complicated reality. When introduced properly and supported by complementary measures, the new bus line can be beneficial. However, without these additional measures the expected improvements may not be realized.

4.5. Comparison of Outcomes across Cities

Table 3. Comparison of key indicators (KPIs) across four cities.

City	Modification	KPIs	Better	Worse
Helsinki	New tunnel	Traffic flow	✓
		Pollution (surface)	✓
		Pollution (tunnel)		✓
Amsterdam	New city quarter	Bicycle traffic flow		✓
		Infrastructure strain		✓
Messina	New bus line	Public transport use	✓
		Private vehicle use		✓
Bilbao	Partial closure	Traffic congestion	✓
	of city center	Emissions	✓
		Traffic spillover		✓

provides a summary of the outcomes following the simulated modifications in each city. In Helsinki, the construction of a new tunnel improves traffic flow and reduces surface-level pollution, but it also significantly increases pollution within the tunnel. In Amsterdam, the addition of a new city quarter is expected to further strain the already overloaded bicycle traffic; however, with targeted adjustments, this impact can be mitigated. In Messina, the introduction of a new bus line shows substantial potential for improvement, especially if it successfully encourages a shift from private cars to public transit. In Bilbao, closing the city center to private vehicles enhances the quality of life within the center, but it also redirects traffic to other areas, requiring a careful balance.

5. Discussion

The implementation of various advanced components and methodologies based on AI and DSS in the context of smart urban mobility projects, such as those in the EU smart city URBANITE project, demonstrates substantial potential advantages tailored to the unique needs and priorities of different urban environments:

• Architecture and components of DSSU: The URBANITE project’s smart urban mobility platform features a sophisticated decision support system, DSSU, that aids decision makers in identifying the most effective urban mobility policies tailored to specific needs and objectives. As an integral part, this system incorporates DEXi, an MCDM modeling software. DSSU assists stakeholders in evaluating a broad spectrum of policy alternatives based on well-defined criteria and performance metrics. By systematically processing these metrics, the system offers data-driven recommendations aligned with strategic urban mobility goals. Its modular architecture ensures ease of updates and enhancements with new data and technologies, maintaining the engine at the forefront of urban mobility solutions. This flexibility and robustness highlight the engine’s ability to adapt to evolving urban mobility challenges.
• Adaptability across urban scenarios: DSSU excels in facilitating MCDA, allowing users to compare multiple policy suggestions or benchmarks by considering an extensive range of objective and subjective key performance indicators. This comprehensive evaluation strategy ensures a well-rounded assessment of policy alternatives. Preference decision models, developed in close collaboration with pilot cities within the URBANITE project, underscore the engine’s adaptability to diverse urban scenarios. The iterative approach of integrating feedback from these pilot implementations enables the system to evolve continuously, ensuring that its results are both effective and contextually relevant.
• Data integration and scalability: The precision and comprehensiveness of input data are critical to the DSSU’s effectiveness. Creating agent plans from the data is essential for the flexibility of the data descriptions. Designed to require minimal user input, the engine reduces the burden on urban planners and decision makers. By analyzing KPIs and other relevant metrics, it generates highly targeted and efficient recommendations, highlighting proposals based on their efficacy. The engine’s scalability is evident in its capacity to handle large volumes of data from multiple sources, making it suitable for cities of various sizes and complexities.
• Exploring the depth of recommendations: DSSU employs a detailed approach to examining the depth of its recommendations, utilizing techniques such as the ±1/2 analysis. This method involves making slight modifications to existing proposals to identify potential improvements or trade-offs. By providing detailed insights into each recommendation, the engine empowers city officials to refine their proposals, gaining a deeper understanding of the potential impacts of different policy alternatives. This analysis ensures that recommended policies are as good as possible and adaptable to future changes and unforeseen challenges. The engine’s ability to provide comprehensive insights builds confidence among stakeholders, promoting a collaborative and informed decision-making process.
• Open-source: DSSU is an open-source research prototype and available for any city to use (https://urbanite-project.eu/).

Various strategies can be integrated to achieve more comprehensive and sustainable urban mobility outcomes such as:

• Enhanced public transport and intermodal transport solutions: Strategies such as increasing the frequency of buses and trams, extending service hours, improving connectivity between different modes of transport, and incentivizing intermodal transport solutions can make public transport a more viable alternative to car travel. This is crucial for addressing the influx of workers commuting by car, which contributes to congestion and pollution. Policies that promote seamless integration of various transport modes, such as public transport, cycling, and walking, can amplify the positive outcomes identified by the DSSU simulations.
• Active transportation infrastructure: Developing and enhancing infrastructure for cycling and walking—including dedicated bike lanes, expanded pedestrian zones, bike-sharing stations, and secure parking at city entrances—encourages the use of sustainable transport modes. These measures reduce dependency on private vehicles and promote healthier, more environmentally friendly travel options.
• Public awareness and engagement campaigns: Launching public awareness and education initiatives is essential for gaining public support and fostering behavior change. These campaigns can increase the adoption of sustainable transportation options and align public perceptions with urban mobility goals. Engaging citizens in the planning process through participatory approaches can also foster greater support and smoother policy implementation.
• Environmental monitoring and adaptation: Implementing real-time environmental monitoring systems to track pollution levels and traffic flow enables cities to adapt their strategies dynamically. This ensures that urban mobility plans remain responsive to changing conditions and can be adjusted to maximize effectiveness.
• Policy and incentive programs: Introducing policies such as congestion pricing, low-emission zones, and incentives for carpooling or electric vehicle use can reinforce the desired outcomes of urban modifications. These policies help reduce traffic congestion, lower emissions, and promote sustainable transportation practices.
• Data standardization and sharing: The effectiveness of DSSU is heavily dependent on the availability and quality of urban mobility data. Policies that promote standardized data collection and sharing across different departments and municipalities, as well as the adoption of open data initiatives and integration of real-time data sources, can significantly improve decision-making processes and the overall functionality of DSSU.
• Regulatory frameworks for sustainable urban mobility: Updating regulatory frameworks to prioritize sustainable urban mobility, such as setting targets for reducing vehicle emissions, encouraging the use of electric vehicles, or implementing congestion pricing schemes, would align with DSSU’s objectives and facilitate the effective implementation of its recommendations.
• Capacity building and training for city planners: To ensure the effective use of DSSU, it is crucial to inform and educate urban planners and decision makers. Training programs focused on the use of DSSU (or similar systems) and related technologies will enable city planners to fully leverage the system’s capabilities and apply its recommendations in a manner tailored to their specific urban contexts.

Overall, the integration of artificial intelligence in decision support systems for urban mobility marks a transformative approach to urban planning, aiming at creating smarter, more sustainable urban environments. As urban centers continue to expand and evolve, the scalability and adaptability of these systems will be critical in ensuring that urban mobility policies remain effective and relevant. Future research should focus on further exploring the potential of these systems, ensuring that cities worldwide can leverage AI to achieve their sustainability goals.

6. Conclusions

6.1. Main Results of the Study

This paper introduces an open-source decision support system (DSS) based on traffic simulations and multi-criteria methodology for evaluating city strategies and policies. The major question addressed is “How will the urban modifications be perceived through objective and subjective KPIs after a specific change is introduced”?

In this study, four European cities—Helsinki, Amsterdam, Messina, and Bilbao—were selected as case studies to evaluate the effectiveness and adaptability of the DSSU system in addressing urban mobility challenges. While these cities were deliberately chosen for their distinct characteristics and varying urban mobility challenges, the comparative analysis of their outcomes reveals both common trends and unique insights that underscore the universality and versatility of the DSSU system.

Despite the differences, some common trends emerged from the application of the DSSU system across all four cities. First, the AI-driven evaluations consistently provided city planners with actionable insights that improved decision making. In each case, the system successfully identified optimal scenarios for traffic management, emission reductions, and infrastructure improvements, proving its capability to handle complex urban mobility issues in diverse environments.

Another shared outcome was the importance of integrating multimodal transportation data into decision-making processes. In all cities, the DSSU system highlighted the need for a holistic approach to urban mobility, where the interactions between different modes of transport (e.g., cars, bicycles, public transit) were crucial in understanding and improving overall traffic flow and emissions. This trend emphasizes the system’s strength in providing a comprehensive analysis that accounts for the interconnected nature of urban transportation networks.

A third joint conclusion drawn from the study was that not all proposed modifications yielded the expected benefits. For instance, while closing a city center to private vehicles reduced direct pollution in the targeted area, it also led to longer travel distances for simulated agents (representing human commuters), which in turn increased pollution in other parts of the city and overall (if commuters did not massively use public transport). Similarly, although constructing a new tunnel successfully reduced surface-level pollution, the concentration of pollutants within the tunnel escalated to a degree that negated the overall improvement in air quality. As observed across nearly all scenarios tested, some KPIs showed improvement while others deteriorated. This outcome underscores the reality that in complex realistic urban mobility planning, there are often trade-offs, and achieving a perfect solution without any negative consequences—often referred to as a "free lunch"—is highly unlikely.

However, the case studies also revealed unique challenges specific to each city, demonstrating the DSSU system’s adaptability to various urban contexts.

• In Helsinki, the primary challenge was managing the traffic congestion around the West Harbour area due to heavy ferry traffic. The DSSU system’s simulations revealed that while the construction of a new tunnel could alleviate surface congestion, it might also concentrate pollution within the tunnel, thus requiring additional measures like improved ventilation or complementary public transit options. This outcome illustrates how the system can identify trade-offs in infrastructure projects, allowing planners to make more informed decisions.
• Amsterdam faced the challenge of integrating a new city quarter without disrupting the city’s extensive bicycle network. The DSSU system’s multi-criteria analysis successfully pinpointed potential bottlenecks in bicycle traffic that could arise from the new development. Enhancements to bike lanes and traffic signal optimization are recommended to mitigate the impact on cyclists. This case underscores the system’s ability to preserve existing sustainable transportation modes while accommodating urban growth.
• Messina’s challenge involved improving public transport connectivity through the introduction of a new bus line. The DSSU system highlighted that while the new bus line could reduce reliance on private vehicles and lower emissions, its success would heavily depend on complementary measures, such as increasing the frequency of bus services and promoting public transport usage. This outcome demonstrated the system’s ability to emphasize the need for a coordinated approach to urban mobility, where infrastructure improvements must be supported by policy and behavioral changes.
• In Bilbao, the focus was on reducing traffic congestion in the city center by partially or fully closing certain streets to private vehicles. The DSSU system identified that partial closures during peak hours could significantly reduce congestion and emissions without causing adverse effects in other areas of the city. However, the system also pointed out that complete closures might lead to traffic spillovers into surrounding neighborhoods, thus requiring a more balanced approach. This case illustrates the system’s capability to propose nuanced solutions that balance the benefits and drawbacks of different policy options.

The comparison across all four cities demonstrates the DSSU system’s ability to deliver data-driven, context-sensitive recommendations, underscoring its potential as a universal solution for managing urban mobility. It effectively addresses specific urban needs while consistently improving sustainability and livability.

AI-driven strategies offer cities advanced tools to identify inefficiencies, anticipate future challenges, and develop innovative solutions. The future of urban mobility will increasingly rely on smart devices and AI, enabling dynamic prediction and adaptation to evolving urban needs. Collaborative efforts among city planners, data scientists, transport experts, and policymakers are crucial to fully harness these strategies, ensuring that cities evolve towards more intelligent and responsive mobility systems.

6.2. Main Scientific Value

The findings from this study provide strong evidence in support of the initial scientific hypotheses. The successful design and implementation of the DSSU system confirm that an AI-based decision support system can be effectively developed both as an algorithmic design and as a fully functional software tool tailored for urban mobility planning. The DSSU system’s ability to manage complex, multi-layered datasets—such as traffic flow metrics, emission statistics, and socio-demographic variables—demonstrates its versatility and adaptability across diverse urban environments. By integrating advanced simulation techniques, including agent-based modeling and multi-objective optimization, the DSSU was able to offer city planners highly detailed, socially aware, and context-sensitive recommendations. These capabilities ensured accurate modeling of real-world scenarios and provided actionable insights, thereby validating the system’s utility as a practical and effective solution.

Moreover, the application of the DSSU system across various European cities not only successfully demonstrated its universality and interoperability but also confirmed that AI-driven evaluations can significantly enhance urban mobility management. In Helsinki, for instance, the system’s simulations of the potential effects of tunnel construction provided critical insights into traffic flow and emissions, effectively highlighting both the benefits and trade-offs associated with such infrastructure projects. Similarly, in Amsterdam, the DSSU’s multi-criteria analysis offered targeted strategies to mitigate the potential negative impacts of a new city quarter on bicycle traffic, ensuring that urban expansion could proceed without compromising the city’s established cycling infrastructure. These case studies underscore how the DSSU system facilitated more informed and effective decision making, leading to direct improvements in the management of urban mobility challenges.

The DSSU system’s use of multi-criteria decision-making frameworks, coupled with AI-enhanced scenario analysis, contributed significantly to improved outcomes in each case study. The system generated optimized scenarios that balanced competing objectives—such as reducing emissions while maintaining traffic efficiency—demonstrating its effectiveness in addressing the complexities of urban planning. Additionally, the inclusion of subjective evaluation criteria, tailored to each city’s specific goals and preferences, ensured that the recommendations were both technically sound and aligned with local priorities.

This comparative approach not only enhances the depth of analysis but also fosters an in-depth understanding of the intricate dynamics involved in smart city development. The DSSU system’s capacity to provide explainable, data-driven insights that guide policy decisions marks a significant advancement in the field of urban planning. The system’s scalability, adaptability, and open-source nature offer a promising solution for diverse urban contexts, contributing valuable knowledge to the ongoing development of smart cities.

Furthermore, the paper identifies future research directions, including investigating the long-term impacts of different policies on urban sustainability, exploring the influence of emerging technologies on the evolution of smart cities, and examining the implications of ambitious policies aimed at achieving zero greenhouse gas emissions. By delineating these avenues for future inquiry, the paper not only contributes to ongoing discussions on smart city development but also lays the groundwork for advancing sustainable urban governance practices through empirical research.

In summary, the research confirms that the DSSU system not only fulfills the initial scientific hypotheses but also provides a powerful tool for navigating the complexities and specifics of modern urban environments. The findings from this study serve as a valuable guide for other cities facing similar challenges, demonstrating that AI-driven decision support systems can significantly enhance the quality of urban life through more informed and effective urban mobility planning.

6.3. Limitations of the Study

In the context of system implementation, one of the foremost challenges pertains to its complexity. The whole URBANITE system comprises approximately two million lines of code, encompassing multiple modules, operating systems, versions, and upgrades. Despite DSSU being comparatively smaller, its functionality relies on intricate data handling and output procedures. Consequently, updates to any module often necessitate integration efforts across various system components. While acceptable as a research prototype, these complexities pose significant hurdles for practical deployment.

Moreover, practical usability is hindered by two additional issues. Firstly, there is a lack of comprehensive manuals, which complicates the system’s adoption and operationalization. Secondly, there existed a notable skills gap among some city team members regarding advanced AI and computer approaches, e.g., multi-objective criteria optimization. The advanced system mandates specialized training before the system can be effectively utilized in practice. Addressing these practical challenges is crucial for enhancing the system’s feasibility and utility in real-world smart city applications.

6.4. Future Research

Future research in the realm of smart city development offers numerous avenues for advancing our understanding and refining the implementation of smart city initiatives, with a heightened emphasis on environmental sustainability and reduced pollution and congestion. One promising area for exploration involves assessing the long-term impacts of advanced AI and decision support system models on urban sustainability, economic growth, and social equity. Conducting “what if” studies to analyze in advance the outcomes of smart city projects enables evaluating the efficacy of diverse strategies in achieving their objectives, and identifying critical success factors and potential pitfalls.

Additionally, there is a pressing need to investigate how these emerging technologies can be seamlessly integrated into existing urban infrastructure and governance frameworks to bolster efficiency, transparency, and citizen engagement. Such insights will prove invaluable for policymakers and urban planners striving to optimize smart city initiatives in real-world settings.

Furthermore, the adoption of radical policies aimed at achieving zero greenhouse gas emissions, particularly focusing on reducing CO₂ emissions, stands to transform the operational landscape of smart cities. As various industrial sectors chart pathways toward net-zero emissions, significant investments in renewable energy sources become imperative to sustain the growing energy demands of smart cities. Overcoming barriers such as economic, social, and administrative constraints, as well as navigating energy crises, will be crucial steps in facilitating the transition to smart energy systems within cities.