Application of Machine Learning and Multi-Dimensional Perception in Urban Spatial Quality Evaluation: A Case Study of Shanghai Underground Pedestrian Street

Abstract

The exploitation of urban subsurface space in urban inventory planning is closely connected to the quality of urban environments. Currently, the construction of underground pedestrian streets is characterised by inefficiency and traffic congestion, making them insufficient for fulfilling the demand for well-designed and human-centred spaces. In the study of spatial quality, traditional evaluation methods, such as satellite remote sensing and street maps, often suffer from low accuracy and slow updating rates, and they frequently overlook human perceptual evaluations. Consequently, there is a pressing need to develop a set of spatial quality evaluation methods incorporating pedestrian perspectives, thereby addressing the neglect of subjective human experiences in spatial quality research. This study first quantifies and clusters the characteristics of underground pedestrian spaces using spatial syntax. It then gathers multidimensional perception data from selected locations and ultimately analyses and predicts the results employing machine learning techniques, specifically Random Forest and XGBoost. The research results indicate variability in pedestrians’ evaluations of spatial quality across different functionally oriented spaces. Key factors influencing these evaluations include Gorgeous, Warm, Good Ventilation, and Flavour indicators. The study proposes a comprehensive and applicable spatial quality evaluation model integrating spatial quantification methods, machine learning algorithms, and multidimensional perception measurements. The development of this model offers valuable scientific guidance for the planning and construction of high-quality urban public spaces.

1. Introduction

According to Lewis Mumford, a harmonious street should be conducive to the free development of human beings and respect human nature [1]. In the context of data-driven urban renewal, spatial governance methods have been further refined [2], and the demand for human-centred and high-quality spatial environments is becoming increasingly urgent. However, as an essential public space in the city, underground pedestrian streets have problems that need to be solved, such as resource waste and traffic congestion, which also lead to opportunities and challenges for spatial quality research.

There are difficulties in the spatial quality evaluation methods for underground spaces, and conventional spatial evaluation methods are not always applicable to such scenarios due to the differences in the air, light, and sound environments compared to above-ground buildings [3].

On the other hand, there is significant potential for improvement in traditional research methods used to evaluate spatial qualities. Currently, most studies rely on data collection techniques such as remote sensing satellite imagery and street view images; however, the timeliness and accuracy of these data sources often hinder in-depth investigations of unique spaces, such as underground pedestrian streets [4]. Regarding data analysis, studies predominantly utilise Space Syntax or Urban Environmental Networks for spatial quantification alongside machine learning for data prediction. Nevertheless, few of these studies integrate spatial quality with human perception, which may conflict with the original intent of street design that prioritises human nature.

To address the challenges mentioned above, this research is conducted in three phases: ①. Quantitative Spatial Analyses, ②. Multidimensional Sensory Perception, and ③. Machine Learning Algorithms. These phases enable the integration and classification of complex spatial environments, yielding multidimensional perception data characterised by varying spatial attributes. Employing multiple machine learning models makes it possible to predict spatial indicators and identify the key indicators influencing spatial quality evaluation.

Therefore, this study introduces spatial quality evaluation models for underground pedestrian streets, integrating machine learning to analyse multidimensional perceptual measures. The model applies to various urban spaces and conducts spatial evaluation research grounded in human multidimensional perception data. Ultimately, it aims to derive indicators that significantly influence spatial evaluation through machine learning algorithms, thereby contributing to the renewal of existing cities and the humane design of future urban environments.

2. Literature Review

In contemporary urbanisation and the continuous evolution of the urban environment, effectively improving the quality of urban space has become a core issue in urban planning and design. As a significant urban catalyst, underground pedestrian streets address the needs of pedestrians, metro traffic, and commercial activities, promoting urban development [5]. Currently, the environmental quality of underground spaces varies across regions, with some areas remaining unused or even abandoned. Consequently, developing effective methods for predicting and forecasting underground pedestrian streets can enhance the design layout of urban spaces to achieve greater efficiency [6]. Yun et al. employed a measurement method for assessing accessibility and the functional layout of underground spaces by selecting spatial syntactic indexes, points of interest (POI) density, and other factors, providing a valuable reference for urban space planning [7]. Gang et al. integrated virtual reality (VR) technology with the spatial scales of underground street spaces and the activities of individuals to explore suitable spatial design methods [8]. However, quantitative models are scarce in urban underground space planning, and the use of multi-source data is infrequent [9]. Furthermore, understanding how the spatial environment influences human behaviour and perception, mainly through human sensory experience data, requires further research.

The increasing demand for high spatial quality among urban residents and the diversity and complexity of underground pedestrian spaces have prompted researchers to examine spatial quality from a more multidimensional perspective. This study employs data-driven quantitative spatial analysis to offer insights into measuring and assessing the structure and function of urban spaces through quantitative methods, thereby establishing a scientific foundation for spatial optimisation. However, quantitative analysis often struggles to fully capture the various impacts of the spatial environment on human perception. To address this limitation, an evaluation of multidimensional sensory perception is introduced to complement quantitative analysis at the micro level, providing a more intuitive reflection of citizens’ actual experiences. Moreover, machine learning enhances the accuracy and efficiency of the assessment, enabling the handling of large-scale, multi-source data and the organic integration of sensory data with quantitative results to create a more comprehensive spatial quality assessment system.

2.1. Data-Driven Quantitative Spatial Analyses

With the advancement of digital technology, the study of urban and architectural qualities has gradually shifted from traditional qualitative analysis to more precise quantitative analysis. This transition has become a hallmark of the modernist movement and parametric design [10]. The investigation of underground pedestrian streets benefits from data-driven quantitative spatial analyses, which effectively evaluate the structure and function of urban environments. Network analysis, a novel approach to urban management, is increasingly utilised in macro urban planning to elucidate the complexities inherent in urban structures and functions [11]. For instance, Zhuang et al. integrated urban road networks with socio-economic characteristics to predict traffic patterns in Chicago [12]. At the same time, Mussone et al. conducted a comparative analysis of underground and bus transport networks through graph theory, highlighting their respective topological characteristics. This work provides a crucial reference for transport planning [13].

OSM (OpenStreetMap) [14] is a collaborative project to create an entirely free and openly accessible world map. Eighty per cent of OpenStreetMap’s global user-generated roadmaps have been completed [15]. Kanthi, N. and other scholars, in the process of mapping an area, have pointed out that the use of OSM dramatically aids in the process of mapping detailed information about the area to data sources in the form of images [16]. The use of OSM can serve as a fundamental reference for spatial quantification. Space Syntax, a spatial quantification tool utilised in architecture and planning, effectively elucidates the interaction mechanisms between individuals and their environments [17]. Furthermore, it analyses the relationship between the degree of spatial integration and environmental perception from a microscopic perspective [18]. Scholars such as Rauof and others have realised specific applications of Space Syntax by using Depthmap X, including measuring connectivity at entrances and exits [19], the walkability of streets [20] and so on. This paper employs spatial syntax as the primary tool to quantify urban spatial quality, especially in studying underground pedestrian streets. It emphasises the importance of understanding how spatial structure influences human perception through these quantitative analyses.

The integration of data technology can effectively improve residents’ quality of life, promote sustainability, and improve public services by integrating and highly integrating information about space [21]. Although spatial syntax offers valuable quantitative data in spatial quality, many scholars primarily focus on quantitative comparisons, often neglecting further applications for in-depth analysis. Relying solely on a single quantitative method and evaluation standard presents inherent limitations [22]. This paper advocates for incorporating human-perceived data into the analytical process to assess urban spatial quality from a human-centred perspective comprehensively. It proposes developing a comprehensive evaluation method integrating spatial quantitative data with perceptual data.

2.2. Spatial Evaluation Based on Multidimensional Sensory Perception

In studying the spatial quality of underground pedestrianised streets, it is insufficient to evaluate their effectiveness solely through quantitative data. True urban spatial quality encompasses structural analyses and multi-dimensional sensory perceptions, which are essential for comprehensively understanding the pedestrian experience within the actual environment. In evaluating spatial quality through multidimensional sensory perception, pedestrians are situated precisely, allowing for a heightened frequency of experiential feelings that conventional remote sensing maps or street view image data cannot replicate. Compared to assessing spatial structure data and image segmentation results derived from Space Syntax, the data obtained through multidimensional sensory perception more closely aligns with human experiences.

Multisensory perception began to gain traction as a significant factor in environmental perception in the latter half of the twentieth century [23]. Lucas stated that urban space should be experienced through all senses, referencing James J. Gibson’s [24,25] theory of perception. They categorise the perceptual system into Vision, Auditory, Olfactory, Somatosensory, Taste, and Kinesthetic, employing a notation system to evaluate urban space [26] comprehensively. However, existing research on multidimensional perception remains theoretical mainly, with a lack of extensive empirical studies.

The perception of street space quality is primarily assessed through subjective evaluation [27], street view images [28], biosensors [29], etc. In selecting indicators, Salesse et al. utilised data from the Place Pulse project to evaluate perceived safety, uniqueness, and other aspects of street space [30,31]. Ewing and Clemente developed a quantitative urban design quality evaluation system by breaking down street space quality into indicators such as neatness, enclosure, and permeability [32]. In the context of underground space, Jeonghwan Kim et al. explore the potential positive effects of artificial windows and plants on the underground environment by evaluating questionnaires and measuring pico-electrical activity [33]. In indoor environmental quality (IEQ) research, there is a gradual progression towards a multi-dimensional understanding of both the acoustic and light environments [34]. While the studies above offer valuable references for methods and indicators related to spatial quality, a systematic evaluation model for spatial quality or a unified set of perceptual indicators has yet to be established. Ying, L. introduced the concept of human-scale urban form in 2016, prompting an increasing number of scholars to adopt a ‘human’ perspective in evaluating spatial quality through sensory perception [35]. Research has primarily focused on the dimensions of Vision [36], Auditory [37], Olfactory [38], Somatosensory [39], and other dimensions have been explored. However, these studies have primarily been conducted from a singular or limited dimensional perspective. The senses should be understood as multidimensional and simultaneous in the process of perception; focusing solely on a limited number of dimensions oversimplifies their impact on spatial experience. Consequently, a systematic approach to processing multidimensional perceptual outcomes is essential. Integrating multidimensional sensory data with spatial syntactic quantification results enhances spatial features. The methodology offers a comprehensive perspective on the various elements involved.

The Literature Analysis Method [40] systematically organises and refines key indicators from existing studies, providing a robust theoretical foundation for the research. This method ensures that the selected indicators possess high academic reliability and accurately reflect scholars’ latest understanding and consensus regarding evaluating spatial qualities. Cui, J. et al. utilised the Literature Analysis Method to distil the core content of their article, analysing the strengths and weaknesses of various underground space cases to offer diverse perspectives on the development and utilisation of underground space [41]. A primary advantage of the Delphi Method [42] is its ability to integrate the expertise and judgment of specialists; through multiple rounds of anonymous consultations, it effectively minimises bias and facilitates the gradual achievement of consensus. Jen-Te Pai et al. employed the Delphi Method to extract criteria, construct an evaluation framework for underground streets, and generate weight values, thereby identifying potential pitfalls in underground pedestrian streets [43]. The above methods can provide a good reference for selecting perceptual indicators for underground pedestrian streets, which is not yet comprehensive.

To investigate the research on spatial quality evaluation through multidimensional perception, this study employs a combination of the Literature Analysis Method and the Delphi Method. Initially, existing multidimensional perception indicators are systematically screened and organised to establish a preliminary indicator framework. Subsequently, the Delphi Method is utilised to optimise these indicators further. This approach not only ensures the comprehensiveness and scientific rigour of the multidimensional perception indicators but also enhances their ability to complement the quantitative data derived from spatial syntax. Ultimately, this results in a more comprehensive and objective spatial quality evaluation model, thereby strengthening the data support for the multidimensional perception outcomes.

2.3. Spatial Evaluation Based on Machine Learning Algorithms

Machine learning models are a powerful tool for managing complex, large-scale datasets. While traditional quantitative analysis methods can yield fundamental insights, they often struggle to accommodate the non-linear characteristics of such data and face limitations when addressing outliers. In contrast, machine learning models enhance the accuracy of spatial quality assessments by thoroughly processing and analysing data from diverse sources. Furthermore, machine learning facilitates a more objective processing and analysis of spatial structure data in spatial quality research, effectively mitigating the cognitive biases that may arise in subjective research.

However, relying only on machine learning may lead to a disconnect from actual human perception, thereby undermining the original intent of exploring the problem from a human perspective. Consequently, it is necessary to integrate multidimensional sensory perception data with objective spatial analysis data and to interpret the results using machine learning models. The application of machine learning models facilitates the effective management of complex datasets and addresses the non-linearity between independent and dependent variables. Several scholars have started employing machine learning models for data processing and prediction in relation to streetscape data [44], urban PM2.5 levels [45], and urban spatial patterns [46].

DBSCAN, Random Forest, and XGBoost have shown significant promise in evaluating urban spatial quality. DBSCAN (Density-Based Spatial Clustering of Applications with Noise), introduced by Ester et al. [47,48], is an unsupervised learning algorithm that excels in handling high-dimensional data due to its density-based clustering approach. Random Forest [49] enhances the model’s generalisation ability while effectively mitigating the risk of overfitting by generating decision trees based on randomly selected datasets. Meanwhile, XGBoost [50] efficiently analyses nonlinear relationships and large-scale datasets, improving prediction accuracy through a weighted voting mechanism. Kisilevich’s research demonstrated the effectiveness of DBSCAN in identifying attractive areas within neighbourhoods by clustering geographic information derived from photographs, thereby validating the reliability of clustering algorithms in neighbourhood-level spatial analysis [51]. Xuening Qin proposed a Random Forest-based model for street environment sensing, utilising high spatial resolution data collected by a mobile sensor network to infer spatial quality at the street level. The study demonstrated the feasibility of employing Random Forests for assessing spatial quality [52]. Meanwhile, Yingji Xia et al. classified traditional features of urban traffic data and introduced a data estimation algorithm based on XGBoost to address anomalous data. Their findings indicated that this approach performs exceptionally well in managing high-dimensional and nonlinear data within urban environments [53]. These studies collectively illustrate that machine learning algorithms exhibit remarkable capabilities in processing various data types within urban spaces. Given the multifaceted nature of the data in this study, the iterative functions of the two models, Random Forest and XGBoost, were employed to optimise the model parameters and enhance accuracy continuously. Ultimately, the results from both models were cross-analysed and validated to yield assessed and predicted values.

2.4. Literature Summary

The research is summarised as follows: In Quantitative Spatial Analyses, the OSM data are utilised for precise spatial mapping. Space Syntax is a reliable tool for analysing spatial structures at both the meso and micro scales, effectively quantifying spatial data. Additionally, the DBCAN algorithm plays a fundamental role in classification through its clustering capabilities. Regarding Multidimensional Sensory Perception, the indexes derived from documentary analysis and the Delphi Method effectively reflect pedestrians’ sensory experiences, yielding more intuitive evaluation results. In the context of Machine Learning Algorithms, the computational features of Random Forest and XGBoost models are employed to efficiently process complex, multidimensional datasets and analyse the relationships between spatial quantification tools and the significance of spatial quality indicators within the framework of Multidimensional Sensory Perception.

All of the aforementioned methods are significant in studying urban spatial quality; however, each possesses certain limitations. While quantitative spatial analyses provide extensive structural and functional assessment data, they often overlook users’ subjective experiences, leading to discrepancies between the findings and actual perceptions. Current research focuses on a single sensory dimension, resulting in an incomplete understanding of sensory interactions. This gap could be addressed through a multidimensional approach to sensory perception. Although machine learning algorithms excel at processing complex data, their reliance on existing datasets and a lack of in-depth understanding of subjective experiences may create a disconnect between theoretical frameworks and practical applications.

In summary, this study will construct a preliminary conceptual model. Initially, quantitative spatial analysis will be employed to assess the spatial structure and function of the underground pedestrian street, providing essential data. Following this, multidimensional sensory perception will be introduced to evaluate users’ experiences within the actual space, complementing the quantitative analysis. Finally, machine learning algorithms will be utilised to integrate the data above and identify complex relationships and patterns, enhancing the capacity of spatial design to predict and optimise user experience. This comprehensive model captures both the objective characteristics of the spatial structure of the underground pedestrian street and a deeper understanding of users’ subjective perceptions. Consequently, it offers more scientific guidance and support for future urban space design and planning, as shown in Figure 1.

3. Materials and Methods

3.1. Case Study

To develop a more representative model for evaluating and analysing the spatial quality of underground pedestrian streets, this study focused on the underground pedestrian street space in the Wujiaochang Yangpu District in Shanghai. Following field research and a review of relevant literature, it was determined that Shanghai, as shown in Figure 2, is a high-density international metropolis located in East China. This city has significant requirements for urban space quality, particularly concerning transport organisation, functional layout, and resource utilisation. Additionally, Shanghai’s underground spaces, developed earlier than in many other cities, currently face challenges such as inadequate design and planning, difficulties in commercial operation, and the inefficient use of space resources in underground construction [54,55]. The Shanghai Urban Master Plan (2013–2035) has outlined requirements for redeveloping existing underground spaces, enhancing the connectivity and coverage of passages surrounding metro stations, and improving the regional non-functioning network [56]. The underground space in the Wujiaochang area of Shanghai has been systematically planned and implemented for over 15 years, contributing to increased urban compactness and enhancing the efficiency of the pedestrian network and the reach of metro services [57]. However, this area continues to experience chaotic phenomena such as the waste of spatial resources, mixed pedestrian flows, and an uneven distribution of businesses. Additionally, research on spatial quality from a perceptual experience perspective remains limited. Therefore, studying underground space in this area is crucial for the city’s renewal and future sustainable development.

From a humane perspective regarding the scope of this study on walking as a mode of transportation, people give up driving and choose to walk a distance of 400 m or 5 min to reach the destination of the distance [58]. The Pongprasert P study indicates that respondents are more receptive to the walking experience when the distance is approximately 500 m [59]. Considering the actual conditions of street coverage and the functional distribution at the specific site of the respondents in this study, the analysis focuses on underground pedestrian spaces within a 250 m radius from the centre of the block where the underground pedestrian street is located. These spaces include underground public pedestrian streets, underground commercial pedestrian streets, and sunken plazas, as shown in Figure 3.

3.2. Research Framework

This study aims to develop a comprehensive evaluation model for the spatial quality of underground pedestrian streets using a multilevel and multidimensional approach. The research methodology integrates Quantitative Spatial Analyses, Multidimensional Sensory Perception, and Machine Learning Algorithms, ensuring that the objectives of each stage are closely aligned with the method, thereby enhancing the scientific rigour and applicability of the model. The research framework is structured into three phases.

①

Phase I: Quantitative analysis of space and feature clustering. Depthmap X and DBSCAN will quantify and analyse the spatial characteristics, determining the positioning of collection points and related parameters within the research area;

②

Phase II: Selection of Multidimensional Sensory Perception Indicators and Data Collection. The Delphi Method and literature review will be utilised for screening questionnaire indicators and data collection, aiming to obtain multidimensional sensory perception evaluation data;

③

Phase III: Determination of indicator weights and impact relationship analysis. The collected data will be weighted and processed, with model predictions conducted using Random Forest and XGBoost. This will ultimately yield the ranking of the importance of perceptual indicators in assessing the spatial quality of underground pedestrian streets, as shown in Figure 4.

3.3. Research Methodology

3.3.1. Quantitative Analysis of Space and Clustering of Features

(1)

Quantitative analysis

Measuring the study area using objective methods that ensure the authenticity and accuracy of spatial data quantification is crucial to conducting quantitative spatial analyses. This study utilises OSM data to compile a comprehensive street and road network dataset, further validated through field research and measurement techniques to ensure data accuracy. Spatial analysis tools are vital in providing the necessary data to support the quantitative examination of spatial structures. This study employs Depthmap X within the Space Syntax framework to analyse the floor plan’s spatial structure and line-of-sight relationships. Additionally, a multi-model analysis method is constructed based on the VGA (Visibility Graph Analysis) and the CAA (Central Axis Analysis) model [60,61].

Depthmap X in Space Syntax was employed to construct the analysis framework, serving as an appropriate scale for assessing pedestrian movement within the area. The VGA model in Depthmap X facilitates the analysis of the spatial structure from a plan view. Integration, Step Depth, and other built-in parameters can be computed and examined. The axis can represent the connection between one public space and another [62]. The CAA model enables the evaluation of each street’s Integration, Connectivity, and Step Depth. This study refers to Xing C.’s method of overlaying indicators [63,64], combining Integration and Step Depth to represent the spatial characteristics of ‘Accessibility’. At the same time, Integration and Connectivity reflect the spatial characteristics of ‘Comprehensibility’. In conjunction with the functional composition of the underground commercial street, these characteristics are categorised into two spatial dimensions: ‘Commercial-oriented’ and ‘Transport-oriented’.

(2)

Feature clustering

Spatial data exhibits multidimensional characteristics, and cluster analysis effectively identifies data clusters with similar spatial attributes, allowing for the classification of clusters with specific spatial features. The DBSCAN algorithm is particularly efficient in processing high-dimensional data and can detect clusters of arbitrary shapes [65]. In this study, the algorithm is employed in a multi-indicator overlay analysis to identify the spatial characteristics of points based on the dimensions of ‘Commercial-oriented’ and ‘Transport-oriented’. After removing noise, DBSCAN successfully locates the distribution of points with distinctive features in the plan view by integrating similar points, thereby providing accurate reference locations for field data collection.

3.3.2. Selection of Multidimensional Sensory Perception Indicators and Data Collection

(1)

Selection of indicators

This study employed the Literature Analysis and Delphi Method to select indicators for multidimensional sensory perception. The Literature Analysis Method evaluates the quality and reliability of existing studies and identifies biases and limitations in the studies. This process enhances the quality of new research and ensures the credibility of the findings. Subsequently, the Delphi Method facilitates further evaluation and refinement of the identified indicators. In this study, the initial indicators screening was conducted using the Literature Analysis Method. Following this, five postgraduate students were organised into a leadership group, and a panel of experts consisting of 24 students and 11 in-service teachers from architecture and town and country planning disciplines were invited to evaluate and refine the indicators.

In the initial round of consultation, experts identified perception indicators pertinent to evaluating spatial quality from existing literature. These indicators were categorised and screened to encompass the dimensions of Vision, Auditory, Olfactory, Somatosensory, Taste, and other recognised international research standards. The resulting indicators include the six principal city perception indicators proposed by Dubey, the Vision dimension [66], and Axelsson et al.’s construct of (PAQs) Perceived Affective Quality of the Auditory dimension [67], etc., culminating in a total of 35 indicators.

In the second round of consultation, the experts reconceptualised and categorised the indicators and shortlist five dimensions comprising 22 indicators, including Security [68], Comfort [69], Beauty [70], etc.

In the final consultation, anonymous feedback was solicited regarding the results of the second consultation until a consensus was achieved. Ultimately, 17 indicators in 4 dimensions—Vision (V), Auditory (A), Olfactory (O), and Somatosensory (S) were selected from the 22 adjective pairs to serve as the evaluation indicators for spatial quality sensory perception in this experiment. Additionally, the questionnaire was designed using a 5-point Likert scale, with ‘Satisfaction’ included as an individual dimension and an overall dimension. For the specific definitions of each indicator, please refer to

Table 1. Spatial quality sensory perception evaluation indicators.

	Spatial Quality Sensory Perception Evaluation Indicators
	Vision perception evaluation	Strongly Disagree	Disagree	Neutral	Agree	Strongly agree
1.	Security	□1	□2	□3	□4	□5
2.	Gorgeous	□1	□2	□3	□4	□5
3.	Non-Repression	□1	□2	□3	□4	□5
4.	Beauty	□1	□2	□3	□4	□5
5.	Interesting	□1	□2	□3	□4	□5
6.	Open	□1	□2	□3	□4	□5
7.	Comfortable Lighting	□1	□2	□3	□4	□5
8.	Vision perception Satisfaction	□1	□2	□3	□4	□5
	Auditory perception evaluation	Strongly Disagree	Disagree	Neutral	Agree	Strongly agree
9.	Quiet	□1	□2	□3	□4	□5
10.	Varied	□1	□2	□3	□4	□5
11.	Pleasant	□1	□2	□3	□4	□5
12.	Auditory perception Satisfaction	□1	□2	□3	□4	□5
	Somatosensory perception evaluation	Strongly Disagree	Disagree	Neutral	Agree	Strongly agree
13.	Good Ventilation	□1	□2	□3	□4	□5
14.	High Wind Speed	□1	□2	□3	□4	□5
15.	Warm	□1	□2	□3	□4	□5
16.	non-humid	□1	□2	□3	□4	□5
17.	Somatosensory perception Satisfaction	□1	□2	□3	□4	□5
	Olfactory perception evaluation	Strongly Disagree	Disagree	Neutral	Agree	Strongly agree
18.	Flavor	□1	□2	□3	□4	□5
19.	Non-pungent	□1	□2	□3	□4	□5
20.	(O)Varied	□1	□2	□3	□4	□5
21.	Olfactory perception Satisfaction	□1	□2	□3	□4	□5
	Overall perception evaluation	Strongly Disagree	Disagree	Neutral	Agree	Strongly agree
22.	Overall Perception Satisfaction	□1	□2	□3	□4	□5

and the detailed descriptions provided in Appendix A.

(2)

Data acquisition

Muleya, N. suggested that to assess the utility of public space quality, it is essential to conduct multidimensional perceptual studies in the field [71]. The present study employed a field distribution of questionnaires to gather multidimensional sensory perception data. This evaluation data were collected through point questionnaires distributed at designated collection points across two dimensions. The questionnaire was organised into three time periods: 11:00–13:00, 14:00–16:00, and 17:30–19:30. During the questionnaire distribution, the Pedestrian Counting method was utilised at the collection points [72]. This method has proven effective for counting pedestrians at entrances and exits, demonstrating accuracy and real-time performance. In this study, 5 min counts were conducted during each period to provide objective data supporting the spatial quality of the underground pedestrian street.

3.3.3. Indicator Weighting and Impact Relationship Analysis

(1)

Indicator weighting

The data collected from the field research collection points were compiled and summarised. The comprehensive weighting method was calculated using the Critic Weighting method. The Entropy Weighting method has been demonstrated to enhance data accuracy in various research types [73,74]. The weights assigned to the identified dimensions were adjusted based on the axial data corresponding to the locations of the collection points. This process resulted in the normalisation of the data, yielding multidimensional perceptual data that encompasses overall spatial feature information. These normalised data were then analysed for multidimensional perception data with distinct spatial characteristics.

(2)

Impact relationship analysis

The various sensory perception data were subjected to Pearson correlation analysis [75], with the Number of Pedestrians serving as a measure of spatial quality to investigate the relationship between each indicator’s influence and spatial quality. Indicators that significantly impact spatial quality were identified and ranked accordingly. Machine learning algorithms, specifically Random Forest and XGBoost, were employed to analyse data regression. Random Forest enhances the model’s generalisation capability by randomly selecting different datasets and generating multiple decision trees. XGBoost efficiently handles non-linear relationships and analyses large-scale datasets. Cross-validation using multiple models ensures that the final analysis is more objective and scientifically robust.

3.4. Subject Situation

The field survey questionnaire collected 236 responses; however, some questionnaires were deemed invalid due to the random selection of participants from passers-by. After filtering, 213 valid data samples remained. Among these, there were 120 male subjects and 93 female subjects. Regarding age distribution, the 18–29 age group had the highest representation with 106 subjects, accounting for approximately 49.78%, followed by the 30–39 age group with 42 subjects, or about 19.72%. The 40–60 age group comprised 28 subjects, representing approximately 13.15%. Fewer subjects were under 18 (20 subjects, 9.39%) and over 60 (17 subjects, 7.98%). This distribution indicates that individuals aged 18–39 are the primary users of the underground pedestrian space. Regarding the overall intentions for using the underground pedestrian street, the proportion of commercial intentions—such as shopping, leisure, and dining—exceeded 41%, while the proportion of transport-related intentions was 38.97%. This suggests that the users’ primary underground pedestrian street space needs are centred around commercial and transport activities.

4. Results

4.1. The Result of Quantitative Analysis of Space and Feature Clustering

Depthmap X was employed to construct the VGA and CAA models, from which the Integration, Connectivity, and Step Depth data were obtained, as shown in Figure 5. The spatial quantitative data from this study were superimposed and analysed using a multi-model computational approach. The intensity of the red colour indicates higher corresponding values. Integrating the two models shows that while the overall numerical trends are similar, distinct results emerge in local areas. For example, in the depth map, the position of the highest depth value in the Step Depth shifted from the VGA model’s northern end to the CAA model’s southern end. This shift may be attributed to the limitations of a single model calculation, which cannot adequately generalise the complexities of the natural environment. In contrast, integrating the two models facilitates a more comprehensive quantification of the spatial structure.

The data derived from the VGA model are classified into ‘Commercial-oriented’ and ‘Transport-oriented’ dimensions. These classifications correspond to the superposition of Integration and Connectivity and the superposition of Integration and Step Depth. The unsupervised autonomous learning method of DBCAN is utilised to extract the point characteristics associated with both the ‘Commercial-oriented’ and ‘Transport-oriented’ dimensions. To present the feature clustering results under these two distinct dimensions more objectively (as illustrated in Figure 6), EPS (Eps-neighborhood of a point) is determined to be 3.8 after several rounds of testing. Following the removal of noise for similar types of integration, the resulting point distribution exhibits significant features In the plan view, which can be utilised for data collection point positioning.

4.2. Questionnaire Reliability Analysis

Combining the Literature Analysis Method and the Delphi Method for indicator selection, four sensory perception dimensions—Vision, Auditory, Olfactory, and Somatosensory—and 17 sensory perception indicators were ultimately identified. The results of the overall dimensions underwent reliability analysis [76]. The Cronbach’s alpha coefficient for the overall questionnaire was 0.912, with a standardised coefficient also at 0.912, indicating the high reliability of the research data. For the reliability analysis of the individual dimensions, the Cronbach’s alpha coefficient for the Somatosensory dimension was found to be 0.697 (as shown in

Table 2. Somatosensory dimension reliability analysis.

Indicator	Corrected Item-Total Correlation (CITC)	The Alpha Coefficient for the Item Deleted Cronbach	Alpha Coefficient
Good Ventilation	0.606	0.572	0.697
High Wind Speed	0.597	0.576
Warm	0.416	0.675
Dry	0.191	0.737
Dissatisfied	0.506	0.623
Standardised Cronbach’s Alpha Coefficient: 0.691

) with a standardised coefficient of 0.691, below the acceptable threshold of 0.7, suggesting average reliability. Notably, the Corrected Item Total Correlation (CITC) for the non-humid indicator was 0.191; after its removal, the Cronbach’s alpha coefficient increased to 0.737 and the standardised coefficient to 0.738, demonstrating improved reliability. Ultimately, the study selected four dimensions and 16 indicators as the essential indicators affecting the quality of the underground pedestrian street for further research.

4.3. Impact Relationship Analysis

4.3.1. Questionnaire Weighting Results

The integrated weighting method was calculated using the critical and entropy weights to derive the four dimensions and the weight relationships among the indicators within each dimension. As shown in

Table 3. Results of the Integrated Weighting Method.

Critic Weighting Method					Entropy Weighting Method			Integrated Weighting Method
Vision dimension	1.081	1.043	1.127	22.977	0.977	0.023	25.822	24.399
Auditory dimension	1.094	1.160	1.269	25.858	0.974	0.026	30.006	27.932
Somatosensory dimension	1.049	1.178	1.236	25.180	0.980	0.020	23.317	24.249
Olfactory dimension	1.021	1.249	1.275	25.985	0.982	0.018	20.856	23.421

, the Auditory dimension exhibits a higher percentage weight than the others. In contrast, the Somatosensory and Vision dimensions display similar rates, while the Olfactory dimension has a relatively low rate. This indicates that the Auditory dimension holds the highest weight among the four dimensions, underscoring its significance in underground pedestrian spaces. Furthermore, the study recalculates the weights of the indicators of each dimension using the comprehensive weight method, revealing that Gorgeous, Quiet, High Wind Speed, and Non-pungent possess the highest weights across the four dimensions, respectively. This study is initiated by quantifying spatial relationships and integrating the CAA model’s Integration, Connectivity, and Step Depth data values to re-evaluate the weights, ultimately producing the final multidimensional sensory spatial quality evaluation data based on the spatial quantification tool.

4.3.2. Correlation Analysis of Multidimensional Sensory Indicators with Spatial Quality Evaluation

This study utilises the spatial quantification tool Depthmap X and a multidimensional sensory perception evaluation method to derive final multidimensional sensory indicator data. A correlation analysis will be conducted with the number of pedestrians, illustrating the spatial qualities. The correlation coefficient matrix of all the obtained multidimensional sensory indicators is shown in Figure 7. The results of the correlation analysis indicate that Vision dimension indicators, such as Open, Interesting, and Gorgeous, exhibit the highest correlation with the number of pedestrians in Commercial-oriented environments. This trend may arise from the convenience pedestrians experience when observing their surroundings in such spaces, leading them to prioritise factors like Openness and Interest during spatial perception. Conversely, indicators such as Gorgeous, Good Ventilation, and (O)Varied in Transport-oriented environments demonstrate the strongest correlation with pedestrian counts. In these settings, pedestrians tend to prioritise Somatosensory and Olfactory perceptions over Vision stimuli to enhance their transport experience. The findings suggest that enhancing the perception of Vision, Auditory, and Somatosensory dimensions significantly influences pedestrian numbers and that strategically improving sensory perception experiences within these spaces can contribute to their overall quality.

To further investigate the high correlation indicators associated with spatial quality, selecting and filtering these indicators for the next phase of the study is essential. In the Commercial-oriented context, the Pleasant, Flavor, and Non-pungent indicators do not significantly affect the number of pedestrian impacts; therefore, they have been excluded from consideration. Similarly, in the Transport-oriented context, the Non-pungent indicator also shows no significant effect on pedestrian influences and has been removed. Ultimately, 13 indicators have been retained for the Commercial-oriented space, while 15 have been selected for the Transport-oriented space for further analysis.

4.3.3. Cross-Validation Analysis of Dual Model Regression

This study employed double-model cross-validation to regress the data to enhance our understanding of how multidimensional sensory indicators affect spatial quality evaluation. The Random Forest and XGBoost models were selected to cross-validate the collated data during the analysis.

Random Forest: Based on the feature importance ranking of the Random Forest dataset after training, it is evident (as shown in Figure 8a–d) that in the Commercial-oriented space, the model performs more favourably on the training set; however, its performance on the test machine is more average, yielding an R² score of 0.549. The most critical feature indicators in this context are Good Ventilation, Warm, Gorgeous, and High Wind Speed. Conversely, in the Transport-oriented space, the model’s performance is more satisfactory on the training set and sound on the test machine, achieving an R² score of 0.699, with Warm, Gorgeous, Good Ventilation, and (O)Varied identified as the essential feature indicators. These findings closely align with the weighting calculations and correlation analyses presented earlier, suggesting a degree of data validity.

XGBoost: Based on the feature importance ranking of the XGBoost dataset after training, it is evident (as shown in Figure 8e–h) that in the Commercial-oriented space, the model demonstrates superior performance on the training set and also performs better on the test machine, achieving an R² score of 0.606. The six essential feature indicators in this context are Good Ventilation, Open, Security, High Wind Speed, Gorgeous, and Warm. Conversely, in the Transport-oriented space, the model exhibits a similarly favourable performance on the training set and achieves a higher R² score of 0.647 on the test machine. This scenario’s seven critical feature indicators include Gorgeous, High Wind Speed, Warm, and Flavour. These findings closely align with the results of the weighting calculations and correlation analyses conducted previously, suggesting that the data possesses a degree of validity.

Cross-Validation Analysis of Dual Model Regression: The cross-validation analysis of the dual model regression results for the Random Forest and XGBoost models indicates that both models yield similar predictions across different domains while demonstrating stable computational performance. In the Commercial-oriented domain, the R² scores exhibit relative smoothness. Conversely, the R² scores are outstanding in the Transport-oriented domain, suggesting that the models effectively generalise the data. Furthermore, both models display a high degree of consistency in crucial feature indicators, which further reinforces the reliability of the analysed data. Specifically, in the Commercial-oriented domain, the critical indicators include Good Ventilation, Open, Security, Warm, Gorgeous, High Wind Speed, Varied, and Quiet. In the Transport-oriented domain, the key indicators are Gorgeous, Warm, High Wind Speed, Flavour, Good Ventilation, Open, Non-Repression, and (O)Varied.

5. Discussion

5.1. Research Innovation

Research on street space is crucial for enhancing citizens’ quality of public life. With the acceleration of urbanisation, the study of street space has diversified significantly. Chen, Yu et al. argue that rapid urbanisation has resulted in health and security challenges in older urban areas. From the perspective of street health security, indicators such as street width, greenery, and nighttime safety severely impact the quality of street space [77]. Wang utilised streetscape images to devise spatial strategies for various street levels [78]. While these studies offer valuable insights from different perspectives, most focus on singular issues or more macro-level analyses of the street. There is a notable lack of research that integrates comprehensive indicators with a more micro-level approach centred on human sensory experiences.

In recent years, some scholars have begun to explore the relationship between space and people from the perspective of sustainable design and sensory experience in street environments. Hu investigated the sustainable design of commercial street space through field surveys and questionnaires [79]. Smeds examined the relationship between the transport space of urban streets and its impact on citizens’ lives [80]. While these studies focus on specific attribute categories of street space, the research frameworks may not universally apply to other types of spaces. Additionally, some scholars have explored the dimension of perception; for instance, Liu et al. focused on Vision perception and analysed how enhancements in Vision treatment can elevate the quality of street spaces [81,82,83]. Nakatani approached the topic from an Auditory perspective, investigating the effects of environmental noise on health and quality of life [84]. Although these studies reveal the relationship between sensory perceptions and spatial environments, there is a notable lack of research on the combined effects of multi-sensory experiences, as most studies focus solely on a single sensory dimension. Therefore, it is imperative to explore in depth how to integrate multiple sensory dimensions for a more comprehensive future assessment in street space research.

This study combines spatial quantification methods, multidimensional perception measures, and machine learning algorithms to provide a richer perspective on spatial analysis by integrating various methods, thereby overcoming the limitations of traditional approaches when addressing complex spatial data. Unlike previous studies, this research ensures the adaptability of the framework by analysing multidimensional data from streets within a 250 m radius of the district and employing objective spatial tools to quantify and categorise street spaces, facilitating the investigation of diverse types of street environments. Field research, questionnaires, and other methodologies were employed to diversify the evaluation of street space from a human-centred perspective of sensory perception, encompassing a broader and more comprehensive range of impact indicators.

In comparison to existing studies, this research introduces novel evaluation indicators derived from multidimensional sensory data, encompassing integrated Vision, Auditory, and tactile perceptions. These new indicators were validated through machine learning models and assigned varying weights. The integration of these multi-perspective indicators offers a more comprehensive reflection of individuals’ subjective perceptions of spatial quality than traditional methods. In previous perceptions, scholars have posited that the human sensory system predominantly relies on five major sensory organs: Vision 83%, Auditory 11%, Olfactory 3.5%, Somatosensory 1%, and Taste 1% [85], with Visual perception being the most dominant in sensory experience. However, in this experiment, the Auditory and Somatosensory dimensions proved to be equally significant, and in some instances, they even surpassed the influence of Vision.

Furthermore, this study found that in commercially oriented underground spaces, traditionally deemed necessary, the ‘Pleasant’ aspect of the Vision experience exhibits a weaker correlation with overall spatial quality. In contrast, Auditory and tactile perceptions significantly influence this quality. This finding diverges from Maffei’s study on Vision and Auditory perceptions [86]. Interestingly, it also supports Aasen’s suggestion of a proven correlation between Olfactory experience and spatial quality [87]. However, the results of M Hedblom’s multidimensional sensory study on urban green spaces [88] indicate that subjects’ Olfactory experiences were more responsive than Auditory and Vision perceptions, which aligns with the current evaluation results of spatial quality in underground pedestrian streets, revealing inconsistencies [88]. The differences in the impact of various sensory dimensions on spatial quality may arise from several factors. Firstly, the physical environments of underground and above-ground spaces differ significantly, leading to distinct sensory feedback. For instance, underground spaces are typically more enclosed, characterised by reduced light and different acoustic conditions compared to above-ground environments. This may result in a more pronounced influence of auditory and tactile senses on spatial experience, while the role of visual perception is comparatively diminished. Secondly, this variability may also be linked to the specific functional orientation of underground spaces, which elevates the importance of non-visual senses in evaluating spatial quality. This phenomenon indicates that the diversity of sensory experiences should be considered when researching underground spaces with varying functional orientations. Furthermore, research methodologies and evaluation indicators should be tailored to the actual use of the space in order to achieve a more comprehensive understanding of the complex relationship between spatial quality and sensory experiences.

This study investigates the specific environments of underground spaces, focusing on the spatial quality of various functional areas. It demonstrates the effectiveness of a multidimensional sensory perception approach across these diverse environments. In conclusion, the study employs innovative research methods and assessment criteria. Multidimensional perceptual measurements facilitate a more comprehensive spatial quality evaluation, particularly in specialised spaces such as underground pedestrian areas. The introduced methodology and new indicators establish a framework for assessing similar spaces. This work enhances the theoretical foundation of spatial quality assessment and provides practical guidance for future urban planning and design.

5.2. Limitations and Future Directions

Although the overall results of this study meet the criteria for each data test, several limitations remain.

In terms of indicator analysis, there are currently more evaluations on the Vision dimension, resulting in a more significant number of Vision indicators than the other dimensions. This imbalance may affect the experimental data somewhat, necessitating further adjustments to enhance the balance of indicator numbers in subsequent experiments. The study also found that the Auditory dimension has the highest proportion in evaluating spatial quality, warranting further exploration into how the relative weights of indicators for each sensory dimension vary with the environment in different functionally oriented urban spaces. In particular, it would be valuable to investigate whether other types of urban spaces (e.g., residential areas, cultural areas, etc.) exhibit different patterns of sensory indicator influence. Additionally, the sense of smell is known to impact human emotions and memories profoundly; however, the Olfactory dimension accounted for the lowest percentage in the evaluation of spatial quality. Future research could investigate why the Olfactory dimension had less influence in this study and whether it may play a more significant role in different cultural contexts or spaces.

In terms of the design of the experimental steps, the study employed a combination of field research and questionnaire analysis to collect data for the sensory perception survey. However, the collection and distribution of the questionnaires were conducted over a brief period, lacking periodicity in testing. This may have overlooked the potential impact of seasonality on sensory perception. Furthermore, focusing exclusively on a single area limits the generalizability of the evaluation. It would have been more compelling if spatial quantification and questionnaire distribution had been conducted simultaneously across multiple regions. Additionally, for future research on sensory perception, recording on-site environmental data would enable a more thorough investigation of the direct impact of objective ecological factors on spatial quality.

This study did not fully capitalise on its high controllability in terms of virtual reality technology application. Virtual reality can be further utilised to enhance the management of independent variables and to investigate the mechanisms influencing spatial quality evaluation. Future research could integrate virtual reality with empirical studies to evaluate spatial quality. The controlled environment of virtual reality can improve the manipulation of independent variables and facilitate a deeper exploration of the mechanisms affecting spatial quality evaluation. Additionally, it is essential to consider how to extend underground research to above-ground spaces using virtual reality technology, thereby examining sensory experiences and their impact on spatial quality across various urban scenarios. This approach would involve simulating sensory experiences in diverse environments and assessing the applicability of different spaces. Ultimately, this research will not only aid in evaluating the quality of existing urban spaces but also provide valuable data to support the simulation and optimisation of new design solutions.

In terms of the application of artificial intelligence techniques, Random Forest and XGBoost have been utilised for model prediction. However, there remains significant potential for enhancing the complexity and granularity of these models when addressing sensory perception data. This study did not fully incorporate deep learning and reinforcement learning models, which could more accurately analyse and predict spatial quality evaluation data, thereby uncovering intricate relationships between sensory data and spatial quality. Future research should consider integrating more advanced AI techniques for predictive modelling. For instance, employing deep learning and reinforcement learning models could facilitate more detailed analysis and prediction of spatial quality evaluation data. This approach would elucidate the complex interactions between various sensory data and spatial quality, ultimately contributing to developing more intelligent and automated urban planning tools.

6. Conclusions

6.1. Conclusions of the Research

By integrating the findings from the correlation analysis and the cross-validation of the Random Forest and XGBoost models, it was concluded that among the two distinct functionally oriented multidimensional sensory indicators of spatial quality, the Auditory dimension accounted for the highest proportion of the four dimensions. The Vision and Somatosensory dimensions were also relatively significant, while the Olfactory dimension exhibited the lowest percentage. The final cross-validated results indicate that the importance indicators—Good Ventilation, Open, Security, Warm, Gorgeous, High Wind Speed, and Varied—were most significant for evaluating spatial quality in Commercial-oriented spaces. In contrast, the indicators Gorgeous, Warm, High Wind Speed, Flavour, and Good Ventilation emerged as the most significant for assessing spatial quality in Transport-oriented spaces.

This study integrates spatial syntactic data with multidimensional sensory data to develop a comprehensive evaluation and analysis model of spatial quality that can be effectively adapted to diverse environments. It employs a multi-model, multi-indicator hierarchical approach. A robust and stable machine learning model is also utilised for feature selection and model predictions. Ultimately, a well-structured and highly suitable spatial quality evaluation system is established.

These research results offer efficient and accurate analysis and evaluation tools for the design and planning of urban public spaces, as well as theoretical and practical support for enhancing the overall environmental quality of these areas. Furthermore, the methodology and findings of this study hold significant importance for the international scientific community, serving as a reference and source of inspiration for further research on urban spatial quality. Future investigations could expand the application scenarios of this model, assess its adaptability in various metropolitan and cultural contexts, and examine the impact of more detailed sensory indicators on spatial quality to foster broader sustainable urban development.

6.2. Suggestion of Design and Application

The correlation and regression analyses of the multidimensional sensory evaluation indicators discussed above can offer valuable insights for the design of current and future pedestrian street spaces:

In Commercial-oriented spaces, greater emphasis should be placed on the Vision, Somatosensory, and Auditory dimensions. Conversely, in Transport-oriented spaces, the focus should shift towards the Vision, Olfactory, and Somatosensory dimensions.

In the Vision dimension, architects should aim to enhance the internal space design of pedestrian streets by incorporating more open spaces, such as atriums or squares, during the preliminary design phase. This approach will help minimise visual obstructions for pedestrians and appropriately increase the height of the space to prevent psychological discomfort. Additionally, it is advisable to consider the integration of outdoor space evaluations in the design of spatial forms, such as assessing the impact of above-ground streets compared to below-ground alternatives. For a completed space, the introduction of natural elements can effectively mitigate visual fatigue; through the thoughtful arrangement of vegetation walls and landscape plants, the overall affinity of the space can be significantly enhanced.

In the Somatosensory dimension, underground spaces possess a distinct advantage over outdoor environments regarding ventilation, wind speed, and temperature control. A study on the refined parameters of indoor air mobility and the frequency of air exchange in control systems is warranted. Additionally, it is crucial to consider indoor temperature conditions and adjust the optimal temperature range to accommodate seasonal variations. The thermal environment can be calculated and analysed by modelling airflow within the space to enhance circulation between indoor and outdoor areas. In built-up spaces, improving indoor air flow and regulating the frequency of air changes within the equipment is essential. If necessary, additional ventilation equipment should be installed to facilitate this process. Furthermore, attention must be given to indoor temperature conditions, which can be adjusted to achieve the most comfortable temperature range according to the season.

In the Auditory dimension, the design process can initiate with dynamic noise control and sound environment design. For instance, during the design phase, it is beneficial to arrange sound-absorbing materials within the necessary walls or to directly install soundproof walls to reduce sound conduction and minimise noise. Additionally, selecting soothing and gentle background music is advisable. Increasing the variety of different tracks can help prevent Auditory fatigue caused by repetitive playback. Furthermore, public facilities, such as posters and chairs, should be regularly replaced to enhance the overall pedestrian experience.

In the Olfactory Dimension, the early planning stage of space design should involve rational design strategies, such as analysing the existing layout of restaurant exhibition spaces and the direction of pedestrian flow. This analysis aims to understand the gravitational effect of pedestrian flow on restaurants within commercial spaces, thereby enhancing the overall pedestrian experience. Odours can be utilised to pique the interest of pedestrians, and through thoughtful flow design, their desire to explore can be stimulated. Furthermore, it is essential to strategically avoid placing rubbish bins, rear kitchen entrances, and other installations or rooms that may emit unpleasant odours in completed spaces. Instead, scenting at main entrances, exits, or characteristic areas can be employed to enhance the space with an ‘Olfactory mark’.

To ensure the successful implementation of these methods in real-world scenarios, municipalities and urban planners should consider the following strategic steps:

①

Pilot Testing in Diverse Neighborhoods: Begin by selecting a few representative neighbourhoods throughout the city that encompass various functional areas, including residential, commercial, and mixed-use zones. In these chosen neighbourhoods, conduct a comprehensive assessment of existing conditions, which should involve data collection on ambient noise, air quality, Vision landscapes, and pedestrian experiences. This approach guarantees that the pilot tests encompass a broad spectrum of urban environments, providing a solid foundation for scaling up the project;

②

Multi-dimensional Sensory Evaluation: Implement a multi-dimensional sensory evaluation system in the pilot neighbourhoods, utilising advanced tools and software for comprehensive analysis. For example, employ sound level meters to quantify noise, gas detectors to assess air quality, three-dimensional (3D) scanning technology to capture Vision landscapes, and thermal imaging to evaluate thermal environments. Furthermore, subjective feedback from residents will be collected through meticulously designed questionnaires encompassing all sensory dimensions, including sight, sound, smell, and touch. This thorough data collection facilitates a holistic understanding of the sensory experiences within urban spaces;

③

Virtual Reality and Simulation for Design Optimisation: Employ virtual reality (VR) and computer simulation technologies to enhance and optimise design before implementation. Create a VR model of the pilot neighbourhoods, allowing citizens to experience the proposed changes virtually and offer immediate feedback. Utilise simulation tools and machine learning techniques to forecast the effects of various design options, including noise propagation and air circulation models. This iterative process guarantees that the final design is practical and attuned to public input;

④

City-wide Policy Development and Innovation Integration: Building on the insights and successes garnered from pilot projects, it is essential to develop city-wide policies and standards incorporating a multi-dimensional sensory evaluation system into urban planning practices. By synthesising data from smart city infrastructure and utilising advancements in AI-driven urban analytics, cities can establish dynamic, real-time sensory evaluation platforms that adapt to evolving urban conditions. Additionally, promoting the development of citizen science programs is essential, enabling residents to actively participate in sensory data collection, which fosters a stronger connection between the community and urban planners. These innovative strategies ensure that the sensory evaluation approach is systematically applied across the city while evolving continuously, integrating the latest scientific and technological advancements to enhance urban livability.