Evaluating Human Settlement Quality: A Novel Approach for Recognizing Feature Importance Based on RBFNN-GARSON

Abstract

The urban living environment is a crucial determinant of the well-being and survival of city residents in modern society. Accurately evaluating the Quality of Human Settlements (QHS) and identifying its influencing factors are essential for advancing sustainable urban development. This study develops an assessment system for the Quality of Human Settlements Index (QHSI) by integrating three dimensions and 29 sub-indicators. The AHP and VIKOR methods are employed to conduct a comprehensive evaluation of the QHS while proposing a factor system of QHS-influencing indicators based on four dimensions and 29 sub-indicators. Additionally, a novel hybrid RBFNN and GARSON method is introduced to identify the importance of QHS-influencing factors. Using Wuhan, China, as a case study, the results reveal that (1) RBFNN-GARSON demonstrates robustness and generalization ability under optimal parameters, effectively revealing the inherent relationships between QHS-influencing indicators and evaluation indicators; and (2) the overall QHSI for Wuhan City is 0.724, with the importance of cultural facilities, historic district preservation, and street cleanliness identified as 0.060, 0.057, and 0.056, respectively, denoting them as the most critical factors influencing the QHS. This method is expected to provide city managers with a deeper understanding of the QHS, enabling them to formulate policies more efficiently, ultimately promoting social progress and enhancing residents’ sense of achievement and happiness. This study contributes to the advancement of sustainable urban development and residents’ well-being.

1. Introduction

The urban living environment is the habitat for a large number of city residents in modern society and its quality is a key indicator for measuring social progress and residents’ well-being [1,2,3]. However, in recent years, the accumulated problems resulting from the overly extensive urbanization process have become increasingly prominent, including increasing urban population density, air pollution, traffic congestion, and ecological degradation [4,5,6], which are also referred to as “urban diseases”, seriously hindering sustainable urban development. For example, a 10 microgram per cubic meter increase in PM2.5 concentration is associated with an approximately 3.8% increase in global mortality [7]. The urban living environment is organically integrated with sustainable development and can promote the efficient operation of various human settlement systems [8,9]. In particular, resident-perceived value refers to the subjective evaluation of the surrounding environment by individuals living in the city, covering multiple aspects such as environment, society, and emotion, and it is a rapid and effective way to improve urban satisfaction. Improving the QHS and resident life satisfaction has become a hot topic of public concern, and a low-quality human living environment not only reduces the willingness of talented individuals to settle in the city but may also pose potential health hazards and bring various hidden costs, thereby weakening the driving force for population inflows [10], which significantly impacts urban development. The QHSI, as a comprehensive evaluation index for the QHS, is a key indicator for measuring resident happiness and satisfaction. Therefore, in-depth research on the evaluation system of resident-perceived value and its influencing factors is of great significance for promoting sustainable urban development and enhancing residents’ well-being.

Urban human settlement directly affects the future development prospects of regions and even countries, and research on the QHS has made significant progress. Yang et al. [11] utilized remote sensing images, in situ meteorological data, the vector data of buildings, and population statistics to seek the optimal layout through nonlinear programming, finding the most comfortable living environment under the urban heat island effect. Luo et al. [12] selected five natural factors and obtained related indicators through batch processing and GIS spatial analysis, utilizing fuzzy comprehensive evaluation to determine weights and proposing an approach to optimize urban spatial form and population distribution. Liu et al. [13] introduced the concept of urban human settlement vitality, measuring the QHS with three elements: urban residents, human environment, and space. They conducted a spatial pattern analysis of human settlement vitality and its influencing factors in four districts of Dalian city using a projection tracking model, spatial correlation analysis, and spatial measure model. Bai et al. [14] constructed a comprehensive evaluation model of urban environmental comfort using multi-source remote sensing data and principal component analysis, analyzing the changes in human settlement environmental comfort from multiple perspectives. Li et al. [15] used five environmental indicators as evaluation criteria and combined analytic hierarchy process and entropy weighting method to comprehensively evaluate urban QHS. Wang et al. [16] developed a comprehensive evaluation system consisting of seven objectives, 23 standard levels, and 34 indicators for the QHS, and provided standard indicator systems based on objective weighting and subjective weighting methods, as well as conducted weight processing of each indicator using the entropy method and the Delphi method. Xie et al. [17] utilized multiple spatial analysis methods to reveal the factors influencing the QHS and systematically analyzed the overall level and spatiotemporal pattern of the QHS in Beijing. Cui et al. [18] selected 18 indicators from three aspects of urban quality of life, production quality, and ecological quality to construct an evaluation index system for the urban QHS in 17 cities and counties in Hubei province, analyzing spatio-temporal differentiation and influencing factors using the entropy weight TOPSIS method and GeoDetector model. Xue et al. [19] established a two-stage model involving spatio-temporal analysis and factor analysis using the Real Human Settlement (RHS) Index and the Pseudo Human Settlement (PHS) Index to analyze the spatio-temporal changes in the suitability and characteristics of urban human settlement environments. Bandauko et al. [20] analyzed the role of a low-quality urban living environment resulting from social unrest, filth, and instability in human development, suggesting that urban decision makers’ attitudes must be more proactive in improving the living environment. Tilaki et al. [21] reconstructed the urban QHS as a reflexive hierarchical structure based on resident data from the northern Iranian capital city of Sari and modeled its satisfaction impact, developing a comprehensive evaluation of the QHS comprising 15 items and a fourth-order reflexive structure. El Nachar et al. [22] proposed a comprehensive model from the bottom-up perspectives of socioeconomic, morphological, and socio-physical aspects, outlining a comprehensive agenda consisting of six analytical tasks. Zhao et al. [23] constructed a human activity index using POI locations, OpenStreetMap road networks, and residential area data, combining machine learning models such as support vector regression, extreme gradient boosting regression, polynomial regression, and random forest regression with remote sensing images to build an urban ecological environment indicator system, revealing the correlation between the street vitality index and mixed index of urban functionality with a sustainable urban ecological environment.

Existing research has made significant contributions to improving the urban QHS and has provided valuable references for future exploration in this field. However, previous studies have often used multi-source data mining or statistical multivariate regression analysis and evaluation models [24]. While these analytical methods generally yield satisfactory results with small sample sizes, achieving reliable and effective results becomes challenging when dealing with large-scale data, especially in subjective cognitive domains [25]. To address these issues, it is necessary to introduce machine learning methods that have undergone extensive training and validation with big data to quantify the importance of the QHS and its influencing factors. These algorithms and statistical techniques offer greater flexibility [26,27,28].

Common machine learning methods for feature extraction include logistic regression (LR), decision trees (DTs), support vector machines (SVMs), and convolutional neural networks (CNNs) [29,30,31,32]. Logistic regression assumes a linear relationship between features and the response variable, which may impose significant limitations on modeling accuracy when the relationship between our features and the response variable cannot be entirely described by a linear function. Decision trees are sensitive to noise and outliers in the data, making them prone to small changes in the data and reducing model robustness, thereby imposing high demands on data quality [33]. An SVM requires significant computational resources when dealing with large-scale datasets, but its efficiency is often not high on these datasets. While a CNN excels in areas such as image processing [34], it fails to fully utilize the internal hierarchical structure of data and does not comprehend the spatial relationships within data in non-image-processing fields. These methods exhibit clear limitations when handling high-dimensional continuous datasets.

In this study, we propose an approach based on a hybrid of VIKOR, RBFNN, and GARSON to achieve feature extraction for the QHS. The VIKOR method is suitable for multi-criteria decision analysis, aiding in the comprehensive evaluation of the trade-offs and relationships between different factors. RBFNN maps low-dimensional linearly inseparable data to high-dimensional space, making it suitable for nonlinear feature extraction and is better at capturing the nonlinear relationships within the data. Compared to other artificial neural networks (ANNs), RBFNN demonstrates strong capabilities in handling high-dimensional data with relatively fewer cluster centers, enabling the more efficient processing of large-scale high-dimensional data while having fewer parameters, making it easier to interpret and understand. The GARSON algorithm helps us to analyze the relative importance of each feature on the output, thereby understanding and explaining the impact of each feature within the model. These methods are widely applied across various fields such as logistics risk assessment, urban building carrying capacity evaluation, and community energy applications [35,36,37].

As one of the most densely populated cities in central China, Wuhan has been selected as a sample city for Chinese urban experiments three times [38,39], making it the focus of this study. This research delves into the inherent correlations between the influencing factors of Wuhan’s urban QHS and the construction of indicators, aiming to provide guidance for urban developers to enhance the QHS. This study accurately evaluates the QHS in Wuhan and determines the importance of each influencing factor, aiming to guide urban developers in improving Wuhan’s QHS, which will contribute to enhancing residents’ well-being and sense of achievement.

The structure of this paper is as follows. In the Section 2, a comprehensive evaluation of the QHS and its influencing factor indicator system is constructed based on a literature review, analyzing the potential inherent relationships between each indicator, and introducing the VIKOR, RBFNN, and GARSON methods used in this research. In the Section 3, these methods are applied to the evaluation of the QHS and the analysis of influencing factors in Wuhan, China, successfully calculating the QHSI and analyzing the importance of each influencing factor. The Section 4 further discusses the computational results of the model. Finally, the Section 5 elaborates on and summarizes the research findings, while also analyzing the shortcomings of this study and outlining future research directions.

2. Materials and Methods

2.1. Materials

2.1.1. Causes of QHS

The urban human settlement environment is the habitat on which a large number of urban residents in modern society rely for survival, and its quality is an important indicator for measuring the progress of human society and residents’ well-being [1]. The perception of urban residents towards the urban environment directly reflects their emotional expressions towards the impact of the urban environment [17], and to a certain extent, reflects the level of life quality evaluation. This perception is often used as an important basis for judging the livability of cities and their future development potential. The sustained attention of the government to this issue and collaborative research with the academic community demonstrates the determination of modern society to pursue sustainable development goals, providing guidance for improving the urban QHS [40]. Due to the multi-faceted nature of residents’ evaluation of the QHS, which includes aspects such as ecological habitability, health and comfort, and convenient transportation, and due to the significant impact of subjective factors in the evaluation process, it is particularly important to establish a comprehensive, objective, and detailed evaluation index system [41]. Based on references from numerous authoritative journals and expert opinions, we have constructed a comprehensive evaluation system for quantifying the QHS, which includes three primary indicators and 29 secondary indicators, as shown in

Table 1. Causes of the QHS indicator system.

Target Layer	Criterion Layer	Scheme Layer	Reference
Human settlement environment perception $\left(A\right)$	$\mathrm{Ecological} \mathrm{habitability} \left(A_1\right)$	Open space $\left(A_{11}\right)$	[6]
		Hydrophilic space $\left(A_{12}\right)$	[6]
		Population density $\left(A_{13}\right)$	[6,17]
		Building height $\left(A_{14}\right)$	[6]
		Park green space $\left(A_{15}\right)$	[42]
		Noise pollution $\left(A_{16}\right)$	[1]
		Air pollution $\left(A_{17}\right)$	[17]
		Water pollution $\left(A_{18}\right)$	[6,17]
	Health and comfort $\left(A_2\right)$	Complete community $\left(A_{21}\right)$	[6]
		Daily nearby shopping $\left(A_{22}\right)$	[40,41]
		Shopping mall $\left(A_{23}\right)$	[6,40]
		Senior dining hall $\left(A_{24}\right)$	[40]
		Inclusive kindergarten $\left(A_{25}\right)$	[40]
		Health service center $\left(A_{26}\right)$	[41]
		Sports venue $\left(A_{27}\right)$	[6]
		Charging pile $\left(A_{28}\right)$	[1]
		Maintenance of infrastructure $\left(A_{29}\right)$	[6,17]
		Activity organization $\left(A_{2A}\right)$	[40]
		Neighborhood relationship $\left(A_{2B}\right)$	[1]
		Housing maintenance level $\left(A_{2C}\right)$	[40]
		Old residential renovation level $\left(A_{2D}\right)$	[40,43]
	Convenient transportation $\left(A_3\right)$	Walking environment $\left(A_{31}\right)$	[42]
		Riding environment $\left(A_{32}\right)$	[42]
		Bus punctuality $\left(A_{33}\right)$	[40]
		Public transport transfer $\left(A_{34}\right)$	[40]
		Railway station setup $\left(A_{35}\right)$	[13]
		Road patency $\left(A_{36}\right)$	[13]
		Parking accessibility $\left(A_{37}\right)$	[6]
		Commuting time $\left(A_{38}\right)$	[40]

. The comprehensive evaluation system of the QHS can help us make more accurate calculations of the QHSI, allowing for a visual and precise quantification of the level of the QHSI in a region. A higher QHSI indicates that residents are living in a high-quality living environment, while a lower QHSI reflects that there are still aspects of the city that need improvement, and the well-being and satisfaction of residents are at a lower level.

2.1.2. Influence Factors of QHS

The perception of the human settlement environment is the individual’s cognitive and emotional evaluation of their life, and it exhibits diversity and individual differences [44]. There are numerous factors that influence the residents’ perception of environmental value. Based on a synthesis of relevant research from authoritative journals in the field, this paper has identified and integrated 29 sub-indicators across four dimensions. These four dimensions are landscape features, pluralism and tolerance, degree of tidiness, and innovative vitality. Landscape features highlight a country’s historical and cultural soft power and also contribute to inspiring modern approaches to enhancing the urban QHS [43]; pluralism and tolerance are essential traits of international metropolises, and understanding and inclusiveness towards migrants and vulnerable groups can enhance the urban QHS [44]; degree of tidiness represents the fundamental appearance of a modern city, because, as a large social system, the city requires effective management and maintenance to maintain order and enhance the QHS [17]; innovative vitality is a key indicator in measuring a city’s future development potential, as innovation is the driving force behind national prosperity and development, and vitality is at the core of urban spirit, so cities lacking innovative vitality often struggle to make residents, especially the younger generation, feel the value of life [8]. These four secondary indicators influence residents’ environmental perception from different perspectives, and each secondary indicator contains several tertiary indicators, which are detailed in

Table 2. Influence of the QHS indicator system.

Target Layer	Criterion Layer	Scheme Layer	Reference
Factors affecting the perceived value of human settlements $\left(B\right)$	Landscape features $\left(B_1\right)$	Landmark building $\left(B_{11}\right)$	[45]
		Cultural facilities $\left(B_{12}\right)$	[43]
		Historical block protection $\left(B_{13}\right)$	[43,45]
		Historic buildings utilization $\left(B_{14}\right)$	[41,43]
		Tourist attraction $\left(B_{15}\right)$	[45]
	Pluralism and tolerance $\left(B_2\right)$	House price acceptability $\left(B_{21}\right)$	[13,46]
		Rent acceptability $\left(B_{22}\right)$	[44]
		Extent of regulation of the housing rental market $\left(B_{23}\right)$	[44]
		Foreign population friendliness $\left(B_{24}\right)$	[45]
		Vulnerable group care $\left(B_{25}\right)$	[44]
		Minimum subsistence level $\left(B_{26}\right)$	[17]
		Safeguarded housing construction $\left(B_{27}\right)$	[46]
		Shantytown rehabilitation level $\left(B_{28}\right)$	[44,45]
		Blind spot $\left(B_{29}\right)$	[44]
		Curb ramp setup $\left(B_{2A}\right)$	[1]
	Degree of tidiness $\left(B_3\right)$	Residential garbage classification $\left(B_{31}\right)$	[40]
		Property management $\left(B_{32}\right)$	[17]
		Street hygiene $\left(B_{33}\right)$	[17]
		Manhole cover maintenance $\left(B_{34}\right)$	[1]
		Vertical pole management $\left(B_{35}\right)$	[46]
		Street lamp management $\left(B_{36}\right)$	[17]
		Parking management $\left(B_{37}\right)$	[1]
		Street sign setting management $\left(B_{38}\right)$	[1]
		Emergency treatment measures $\left(B_{39}\right)$	[1]
	Innovation vitality $\left(B_4\right)$	Talent introduction policy $\left(B_{41}\right)$	[40]
		Job opportunity $\left(B_{42}\right)$	[40]
		Market environment $\left(B_{43}\right)$	[41]
		Scientific and technological innovation environment $\left(B_{44}\right)$	[40]
		Youth appeal $\left(B_{45}\right)$	[10]

.

2.1.3. Relationship between QHSI and Impact Indicators

Based on the comprehensive evaluation of the QHS constructed in this paper, we analyzed the inherent interaction relationships between the QHS and its influencing factors, as illustrated in Figure 1. We have identified four main influencing aspects:

(1)

Landscape features exert a significant impact on enhancing a city’s ecological livability. They not only reflect a region’s profound historical and cultural heritage, increasing the city’s cultural charm and attractiveness to attract more talent and investment, but also enhance the city’s cultural atmosphere and artistic taste, promoting the prosperous development of culture and art, thereby improving the overall livability of the city [47]. For instance, exceptional cultural facilities with distinctive architectural character foster a heightened sense of environmental vigilance and awareness among residents. Residents’ involvement in urban management creates a green environmental atmosphere, increasing the city’s waterfront space and population density, which in turn affects ecological livability and the QHS.

(2)

Pluralism and tolerance play a pivotal role in the QHS in various ways. This inclusive social attitude makes our society more like a whole. Caring for special groups, managing tactile paving, and other measures can reduce internal conflicts and promote understanding and inclusiveness among community members, building harmonious relationships in the neighborhood, thereby improving the comfort and health of the community [48]. Additionally, effective control systems for housing prices, rental rates, and welfare for migrants have a significant impact on the settlement of non-local populations, enhancing the city’s livability [49].

(3)

The degree of tidiness is closely intertwined with the urban human settlement environment. The implementation of measures such as community waste sorting and street hygiene management implies better community sanitation, higher air quality, and lower noise pollution, providing residents with a more comfortable living experience [50]. Strengthening the management and maintenance of streetlights and roadworks aids in constructing integrated communities, resulting in healthier and more comfortable community living for residents. Furthermore, elevating the level of streetlight management and maintenance reduces the possibility of road congestion, rendering transportation more convenient for residents [51].

(4)

The innovative vitality of a city serves as a catalyst for promoting convenient transportation. A community with high innovative vitality indicates a more robust market environment and technological innovation environment, attracting more talent and investment. Such a community can introduce more intelligent and user-friendly transportation systems, enhancing transportation convenience and safety, reducing traffic congestion and the occurrence of traffic accidents, and ultimately improving residents’ travel experiences [52].

In this study, we utilized the Dell PowerEdge R940 server, located in TX, USA, for computational tasks, alongside the Synology DiskStation DS918+ from Synology Inc., based in Taipei, Taiwan, for data storage and management. These devices effectively fulfilled our requirements for high-performance computing and secure data management, which were crucial for our research. Additionally, various software tools were employed for data analysis, chart creation, and spatial data management. Specifically, Python 3.12.1 (Python Software Foundation, Wilmington, DE, USA) was used for data manipulation and scripting; Microsoft Visio 2021 (Microsoft Corporation, Redmond, WA, USA) facilitated the creation of charts and flow diagrams; statistical analysis and data visualization were performed using Origin 2024 (OriginLab Corporation, Northampton, MA, USA); and geographic information system (GIS) analyses were conducted with ArcGIS 10.8 (Environmental Systems Research Institute, Redlands, CA, USA).

2.2. Methods

This paper proposes a multi-indicator evaluation system for the feature extraction of urban human settlement environment influencing factors. The model consists of three parts, with the specific process illustrated in Figure 2. In the establishment of the initial assessment matrix and based on the determination of indicator weights using the AHP method, the VIKOR algorithm is utilized to effectively evaluate the QHS. Subsequently, the influencing factor indicators of the QHS serve as input features, and the RBFNN is employed for environmental quality prediction, achieving the effective integration of performance parameters. Finally, the GARSON algorithm is utilized to scrutinize the importance of the constructed prediction model, revealing the inherent interaction relationships between the QHS and its influencing factors, thereby achieving effective feature extraction. The proposal of this model provides a new analytical tool for the evaluation and optimization of the QHS.

2.2.1. VIKOR Method

The VIKOR (VlseKriterijumska Optimizacija I Kompromisno Resenje) algorithm, also known as the multi-criteria compromise solution method, was proposed by Serafim Opricovic and Gongzhi Fan in 1998 [53]. It is a multi-criteria optimization method based on the proximity of the assessment values of the options to the ideal solution. The core concept of the VIKOR method involves comparing the values of group utility, regret, and comprehensive utility, and to conduct compromise processing on the evaluated options, thereby ranking the merits of the options and selecting the best compromise solution closest to the ideal solution. The advantage of the VIKOR algorithm is its ability to handle complex decision-making problems involving multiple criteria and various types of data, particularly in managing the non-comparability of criterion weights and ideal solutions with actual data. Currently, the VIKOR method has been widely applied to multi-criteria decision-making problems. The specific calculation steps are as follows.

STEP1 Establish the initial evaluation matrix. Assuming there are m evaluation objects in the evaluation system, each with $n$ evaluation criteria, this forms an $m\mathrm{*}n$ initial matrix A:

$$A=\left(\begin{array}{cccc}{a}_{11}& a_{12}& \cdots & a_{1n}\\ a_{21}& a_{22}& \cdots & a_{2n}\\ ⋮& ⋮& \ddots & ⋮\\ a_{m1}& a_{m2}& \cdots & a_{mn}\end{array}\right)$$

(1)

where $a_{ij}$ represents the value of the $j-th$ evaluation criterion of the $i-th$ evaluation object.

STEP2 Matrix normalization. The initial evaluation matrix A is subjected to dimensionless processing to yield the normalized matrix B.

For the benefit-type indicators, the following Formula (2) can be utilized for calculation.

$$b_{ij}={\displaystyle \frac{a_{ij}-\underset{j}{\min }{a}_{ij}}{\underset{j}{\max }{a}_{ij}-\underset{j}{\min }{a}_{ij}}}$$

(2)

For cost-type indicators, the following Formula (3) can be utilized for calculation.

$$b_{ij}={\displaystyle \frac{\underset{j}{\max }{a}_{ij}-a_{ij}}{\underset{j}{\max }{a}_{ij}-\underset{j}{\min }{a}_{ij}}}$$

(3)

The normalized matrix B can be represented as shown in Equation (4):

$$B=\left(\begin{array}{cccc}{b}_{11}& b_{12}& \cdots & b_{1n}\\ b_{21}& b_{22}& \cdots & b_{2n}\\ ⋮& ⋮& \ddots & ⋮\\ b_{m1}& b_{m2}& \cdots & b_{mn}\end{array}\right)$$

(4)

where $b_{ij}$ is the dimensionless form of $a_{ij}$.

STEP3 Calculate the positive and negative ideal solutions $x_j^{+}$ and $x_j^{-}$ as shown in Equations (5) and (6).

$$x_j^{+}=\underset{j}{\max }{b}_{ij}$$

(5)

$$x_j^{-}=\underset{j}{\min }{b}_{ij}$$

(6)

STEP4 Calculate the group utility value $S_i$ and individual regret value $R_i$ for each evaluation object, as shown in Equations (7) and (8):

$$S_i={\displaystyle \sum _{j=1}^n{w}_j\frac{x_j^{+}-b_{ij}}{x_j^{+}-x_j^{-}}}$$

(7)

$$R_i=\underset{j}{\max }{w}_j{\displaystyle \frac{x_j^{+}-b_{ij}}{x_j^{+}-x_j^{-}}}$$

(8)

where $w_i$ is the weight of the evaluation object for each evaluation criterion.

STEP5 Calculate the benefit value $Q_i$ for each assessment object, as shown in Equation (9):

$$Q_i=\lambda {\displaystyle \frac{S_i-\min S_i}{\max S_i-\min S_i}}-(1-\lambda ){\displaystyle \frac{R_i-\min R_i}{\max R_i-\min R_i}}$$

(9)

where λ is the decision-making mechanism coefficient, and λ ∈ (0,1). When λ > 0.5, it indicates that the decision maker leans more toward group utility. Conversely, when λ < 0.5, it indicates that the decision maker leans more toward individual regret. Meanwhile, when λ = 0.5, it indicates that a balanced compromise approach is adopted for decision making.

STEP6 The decision credibility verification establishes constraints as shown in Equations (10)–(12):

$$Q_{\min 2}-Q_{\min 1}\ge {(m-1)}^{-1}$$

(10)

$$S_{\min 1}<S_{\min 2}$$

(11)

$$R_{\min 1}<R_{\min 2}$$

(12)

where $min1$ and $min2$ are the subscripts that minimize and second minimize the value $Q_i$, respectively. If Equations (10)–(12) are simultaneously satisfied, then the decision corresponding to $Q_{min1}$ is considered the optimal solution. If only Equation (10) is satisfied, then the decisions corresponding to both $Q_{min1}$ and $Q_{min2}$ can be considered optimal solutions. If only Equations (11) and (12) are satisfied, then all decisions that satisfy $Q_i-Q_{min1}<{(m-1)}^{-1}$ can be considered optimal solutions [54].

2.2.2. RBFNN-GARSON Method

The RBFNN is a type of single-hidden-layer feedforward artificial neural network (ANN) based on function approximation, proposed in the late 1980s. The core idea of this network lies in the utilization of radial basis functions, which are real-valued functions that depend only on the distance from the origin. The fundamental concept of radial basis functions is to map low-dimensional linearly inseparable data into a high-dimensional space, thereby rendering it linearly separable in the high-dimensional space [55]. There are various types of artificial neural networks, among which the RBFNN, due to its simple structure, excellent generalization ability, strong nonlinear approximation capability, and rapid convergence speed, has been widely applied in various big data analyses.

One of the most distinctive features of RBFNN is its three-layer feedforward neural network structure, with only one hidden layer, as shown in Figure 3. In the figure, $x={[x_1,x_2,x_3,\dots ,x_n]}^T$ represents the input layer signal, $h={[h_1,h_2,h_3,h_4,\dots ,h_m]}^T$ denotes the hidden layer signal, and the final output layer signal is the linear weighted sum of the hidden layer output signals. Compared to traditional ANNs, RBFNN demonstrates its unique advantages in dealing with complex data, as illustrated in the working mechanism of typical neurons in Figure 4.

The specific calculation process of RBFNN is as follows.

STEP1 Set initial values for the parameters of the neurons. The initial values for the central position vector of the basic functions $c_1$, the radius of the hidden layer neurons $d_i$, and the connection weights $w_i$ between the neurons and the output layer are determined as shown in Equations (13)–(15):

$$c_i={\displaystyle \frac{1}{N}}{\displaystyle \sum _{j=1}^M{x}_{ij}}$$

(13)

$$d_i={\displaystyle \frac{c_{\max }}{\sqrt{2h}}}$$

(14)

$$w_i=\exp (-{\displaystyle \frac{h}{c_{\max }^2}}||x_i-c_i|{|}^2)$$

(15)

where $N_i$ represents the total number of samples in the $i-th$ group, M represents the total number of elements in the $i-th$ group, and $x_{ij}$ represents the $j-th$ sample in the $i-th$ group.

STEP2 Train the network using the gradient descent method. The three aforementioned parameters are then iteratively updated until the network achieves the minimum error, following the iteration method described in Equations (16)–(18):

$$c_{ji}(t)=c_{ji}(t-1)-\beta {\displaystyle \frac{\partial F}{\partial c_{ji}(t-1)}}+\alpha [c_{ji}(t-1)-c_{ji}(t-2)]$$

(16)

$$d_{ji}(t)=d_{ji}(t-1)-\beta {\displaystyle \frac{\partial F}{\partial d_{ji}(t-1)}}+\alpha [d_{ji}(t-1)-d_{ji}(t-2)]$$

(17)

$$w_{ji}(t)=w_{ji}(t-1)-\beta {\displaystyle \frac{\partial F}{\partial w_{ji}(t-1)}}+\alpha [w_{ji}(t-1)-w_{ji}(t-2)]$$

(18)

where t represents the iteration number, α and β are the learning factors, and F represents the objective function [56].

STEP3 Compute the activation function of the hidden layer units. The output layer units are computed as shown in Equation (19).

$$R_i(x)=\exp (-{\displaystyle \frac{\left|\right|x_i-c_i|{|}^2}{2d_i^2}})$$

(19)

STEP4 Compute the output layer units as shown in Equation (20):

$$y={\displaystyle \sum _{i=1}^s{w}_i}{R}_i(x)$$

(20)

where s is the number of nodes in the hidden layer.

The GARSON algorithm is a method utilized for the sensitivity analysis of connection weights in neural networks, determining the importance of each input unit by evaluating its impact on the model output. Within the framework of the GARSON algorithm, the contribution of input units to output units is measured by examining the connection weights between the hidden layer neurons and the output layer, and allocating them to the respective input signal’s connection weight [57]. This approach provides a quantitative description of the relative importance of input signals, aiding in understanding their role and priority within the model. The specific method is detailed in Equation (21):

$$P_i={\displaystyle \frac{{\displaystyle \sum _{j=1}^L(|w_{ij}{w}_{jk}|/{\displaystyle \sum _{k=1}^M|w_{jk}|)}}}{{\displaystyle \sum _{i=1}^N{\displaystyle \sum _{j=1}^L(|w_{ij}{w}_{jk}|/{\displaystyle \sum _{k=1}^M|w_{jk}|)}}}}}$$

(21)

where $P_i$ represents the relative importance of the input signal, and $w_{ij}$ and $w_{jk}$ are the connection weights between the input hidden and hidden output layers, respectively.

Statistical error analysis constitutes a crucial method for evaluating the predictive performance of models. To comprehensively assess the effectiveness of our constructed model, this study employs three regression evaluation metrics, namely Mean Squared Error (MSE), Coefficient of Determination (R²), and Mean Absolute Percentage Error (MAPE). Applying these metrics not only quantifies the predictive accuracy of the model but also enables performance comparison among numerous models, thereby providing clear directions for further model optimization. Without validating these metrics, we cannot accurately assess the predictive performance of the model, nor can we identify potential avenues for model improvement.

$$MSE={\displaystyle \frac{1}{N}}{\displaystyle \sum _{i=1}^N(Y_i}-Z_i{)}^2$$

(22)

$$R^2=1-{\displaystyle \frac{{\displaystyle \sum _{i=1}^N(Y_i}-Z_i{)}^2}{{\displaystyle \sum _{i=1}^N(Y_i}-\stackrel{\_\_}{Y_i}{)}^2}}$$

(23)

$$MAPE={\displaystyle \frac{1}{N}}{\displaystyle \sum _{i=1}^N|\frac{Y_i-Z_i}{Y_i}}|\times 100\%$$

(24)

Within the context, $Y_i$ and $Z_i$ represent the corresponding actual and predicted values. The Mean Squared Error (MSE) calculates the absolute value of the squared difference between the predicted and actual values, and a lower standard error indicates higher model accuracy. R² measures the proportion of the variability explained by the model relative to the total variability. It is commonly employed to assess the model’s fit to the data. A value closer to 1 indicates better model fitting, while a lower value suggests the need for improved model fit. The Mean Absolute Percentage Error (MAPE) represents the average percentage error between actual and predicted values. Expressed as a percentage, it is suitable for models with different scales. A MAPE of 0% represents a perfect model, while a value exceeding 100% indicates a poor model [58]. These metrics collectively provide a scientific basis for evaluating and optimizing the model.

3. Results

3.1. Study Area

The city of Wuhan (located at 113°41′ to 115°05′ east longitude and 29°58′ to 31°22′ north latitude), situated at the confluence of the Yangtze River and the Han River, has a specific location, as shown in Figure 5. As the capital of Hubei Province, Wuhan is a sub-provincial city, a central city, and a mega-city in China. As a core city in central China, Wuhan is not only an important industrial base nationwide but is also a vital hub for education, science, and transportation [59]. The whole city encompasses 13 districts with a total area of 8569.2 square kilometers. By the end of 2022, the permanent population in the urban area reached 13.79 million, an increase of 90,100 from the previous year. Notably, the permanent urban population was 11.631 million, accounting for 84.7% of the total population, with an urbanization rate increasing by 0.1% from the previous year. In 2023, the regional GDP of Wuhan reached 2001.17 billion yuan, an increase of 5.7% over the previous year, ranking the city ninth among 293 prefecture-level cities in mainland China. Wuhan stands as the largest city in the middle reaches of the Yangtze River Urban Agglomeration [60].

In recent years, despite the increasing urbanization level and rapid economic and social development in the Wuhan metropolitan area, residents’ satisfaction with their living conditions has not undergone a synchronous improvement. On the contrary, issues such as resource scarcity, environmental pollution, and frequent disasters have grown increasingly prominent due to unrefined urban development. From 2014 to 2023, Hubei Province consistently ranked among the top in the number of traffic crime cases in China, with the province leading the nation in traffic accidents in 2023 [61]. These trends underscore the conflicts between the urbanization process and the living environment [62]. Furthermore, at the beginning of 2020, Wuhan was impacted by the novel coronavirus (COVID-19), further highlighting the increasingly sensitive and acute issue of balancing public health and urbanization [63]. Although the Wuhan municipal government has implemented a series of policies, such as the four-tier “Lake Chief” management system for water environment management, the “one point, one policy” principle for air pollution control, and the “Embrace Blue Sky” campaign, the effectiveness of these measures has not been significant [64,65]. Hence, selecting Wuhan as the focal point of this study is highly representative and typical, given its circumstances.

3.2. Data Processing

The data used in this study were derived from the 2023 Wuhan Urban Health Survey Questionnaire [66]. Drawing on the constructed evaluation index system and influencing factor system, we evaluated the QHS from three dimensions: ecological habitability, health and comfort, and convenient transportation. Additionally, we assessed the influencing factors of the QHS from four dimensions: landscape features, pluralism and tolerance, degree of tidiness, and innovative vitality. The survey participants, residents of Wuhan, rated 58 specific indicators on a satisfaction scale ranging from 0 to 100. A higher score indicates a higher level of satisfaction with the corresponding indicator. A total of 10,146 valid questionnaires were collected in this survey, and after rigorous quality checks and screening, 9913 sets of valid data were obtained. The distribution of scores for the QHS evaluation indicators is presented in Figure 6, while the distribution of scores for the influencing factor indicators is shown in Figure 7. In the figures, the different colors on the vertical axis represent the varying levels of satisfaction of the residents with the corresponding indicators on the horizontal axis.

According to Figure 7, the median satisfaction scores for the majority of evaluation indicators of the QHS in Wuhan City are clustered around 80, reflecting a high overall acceptance of the urban environment by residents. However, indicators such as building height ($A_{14}$), daily nearby shopping ($A_{22}$), senior dining hall ($A_{24}$), and railway station setup ($A_{35}$) have median scores ranging from 40 to 60. Particularly noteworthy is the fact that 1353 surveyed residents rated daily nearby shopping ($A_{22}$) below 20, highlighting significant issues in the convenience of daily shopping in certain areas of Wuhan. Furthermore, among the influencing factors of the QHS, the mode scores of almost all indicators concentrate in the range of 80 to 100. However, more than 20% of the surveyed residents rated parking management ($B_{37}$) and youth appeal ($B_{45}$) below 40, while 1025 surveyed residents rated the restoration and utilization of historical narratives and traditional residences ($B_{14}$) below 20. These indicators have brought to light important factors constraining the improvement of the QHS in Wuhan.

3.3. Score Distribution Characteristics of QHS

In order to conduct an objective and accurate calculation of the indicator weights, we invited 17 experts to be interviewed, including 10 urban planning experts and 7 university experts. All of these experts have lived in Wuhan for over 15 years and have over 10 years of experience in urban planning, community management, infrastructure construction, promotion of civil behavior, and urban environmental protection. Prior to inviting experts to score, we established clear evaluation criteria and indicators to ensure that the experts understood the scope and standards of the assessment. Additionally, during the AHP scoring process, all experts conducted the assessments anonymously and independently to ensure that their opinions were not influenced by others [67]. Through consulting the aforementioned 17 experts, we completed the construction of the AHP weight matrix for 29 QHS evaluation indicators, with specific assessment matrices and consistency test results detailed in Appendix A. The analysis of Appendix A reveals that the Consistency Ratio (CR) values of all constructed weight matrices are less than 0.1, indicating the successful passing of the consistency test. The final calculated weights of each variable are presented in

Table 3. Weight of the QHS indicators.

Criterion Layer	Scheme Layer	Weight
$A_1$ (0.261)	$A_{11}$	0.117
	$A_{12}$	0.067
	$A_{13}$	0.172
	$A_{14}$	0.030
	$A_{15}$	0.073
	$A_{16}$	0.263
	$A_{17}$	0.238
	$A_{18}$	0.040
$A_2$ (0.411)	$A_{21}$	0.046
	$A_{22}$	0.080
	$A_{23}$	0.024
	$A_{24}$	0.050
	$A_{25}$	0.148
	$A_{26}$	0.107
	$A_{27}$	0.018
	$A_{28}$	0.036
	$A_{29}$	0.122
	$A_{2A}$	0.153
	$A_{2B}$	0.050
	$A_{2C}$	0.036
	$A_{2D}$	0.131
$A_3$ (0.328)	$A_{31}$	0.065
	$A_{32}$	0.151
	$A_{33}$	0.124
	$A_{34}$	0.203
	$A_{35}$	0.041
	$A_{36}$	0.188
	$A_{37}$	0.185
	$A_{38}$	0.042

.

From a macro perspective, health and comfort ($A_2$) has the greatest proportional contribution to the composition of the QHS, indicating that the level of community health and comfort ($A_2$) most significantly reflects the overall magnitude of the QHS, followed by convenient transportation ($A_3$), with ecological habitation ($A_1$) having the smallest weight. From a micro perspective, noise pollution ($A_{16}$) and air pollution ($A_{17}$) each account for over 23.0% in the composition of ecological habitation ($A_1$), while building height ($A_{14}$) and water pollution ($A_{18}$) each account for less than 5.0%. In the composition of health and comfort ($A_2$), activity organization ($A_{2A}$) holds the highest weight at 15.3%, while shopping mall ($A_{23}$) and sports venue $\left(A_{27}\right)$ have only around a 2.0% weighting. Regarding convenient transportation ($A_3$), public transport transfer ($A_{34}$) holds a weight of 20.3%, with railway station setup ($A_{34}$) and commuting time ($A_{38}$) displaying the lowest weights at approximately 4.0% each.

Using the comprehensive evaluation of the QHS constructed earlier and the corresponding weights of each indicator, this study utilized the collected data on QHS-influencing factors as input. Each set of resident evaluations was treated as a single scenario, and the VIKOR method was applied to calculate the group utility values, individual regret values, and benefits for ecological habitation ($A_1$), health and comfort ($A_2$), and convenient transportation ($A_3$), with their average values listed in

Table 4. Analysis results of the VIKOR method on the QHS indicators.

	S_Mean	R_Mean	Q_Mean
EHI	0.233	0.068	0.243
HCI	0.262	0.046	0.272
CTI	0.277	0.080	0.289
QHSI	0.259	0.025	0.274

. Recognizing that the QHS leans toward group utility values, the decision coefficient λ in this context was set at 0.8. By further comparing benefits and the distance from the optimal solution, the Ecological Habitat Index (EHI), Community Health Comfort Index (HCI), Transportation Convenience Index (CTI), and QHSI were derived. The results were statistically analyzed, as shown in Figure 8, and the results of various districts in the central urban area of Wuhan City are presented, as detailed in Figure 9.

In the analysis of Table 4, we observed that, among the three indices EHI, HCI, and CTI, the average S, R, and Q values for CTI are all the highest, while EHI presents the lowest values. This suggests that the ecological habitation ($A_1$) in Wuhan City not only receives high recognition from the majority of residents but also exhibits reasonable scores even in the lower range, which is contrary for convenient transportation ($A_3$). Based on the comprehensive analysis of these four indices, the ranking based on average values from high to low is EHI, QHSI, HCI, and CTI.

In Figure 8, the horizontal axis represents the four index scores and the vertical axis represents the corresponding score distribution. The upper and lower boundaries of the box symbolize the upper and lower quartiles of the respective indices, with the length of the box representing the interquartile range (IQR). The line within the box represents the median, the square inside the box represents the mean, and the lines outside the box represent the upper and lower edges. It can be observed from the figure that the mean of each box is above 0.7, reflecting a relatively high level for QHSI, EHI, HCI, and CTI in Wuhan. Within EHI, HCI, and CTI, the median and mean of the EHI are the highest, reaching 0.759 and 0.757, respectively, indicative of a relatively good ecological livability in Wuhan. Additionally, the IQR of EHI is the largest, reaching 0.247, suggesting significant differences in residents’ perceptions of ecological habitation ($A_1$). The median and mean of HCI and CTI are almost identical, while the IQR of CTI, at 0.236, is greater than the IQR of HCI, which stands at 0.217.

Figure 9 displays the four QSI-related indices for the seven districts in the central urban area of Wuhan, namely Jiang’an District, Hanyang District, Wuchang District, Qingshan District, Qiaokou District, Hongshan District, and Jianghan District. The color depth in the figure reflects the respective mean values of these indices for each district. By analyzing Figure 9a, it can be observed that the EHI mean value is highest in Jianghan District, reaching 0.774, whereas Hongshan District has the lowest EHI mean value at only 0.722. Upon further observation of Figure 9b–d, it can be noted that Wuchang District has the highest mean values for HCI, CTI, and QHSI, reaching 0.739, 0.736, and 0.746, respectively. Conversely, Hongshan District and Qiaokou District rank at the bottom two positions. Qiaokou District has a CTI mean value of 0.677, slightly lower than Hongshan District’s 0.679, and Hongshan District’s HCI and QHSI mean values are 0.692 and 0.696, both lower than Qiaokou District’s 0.695 and 0.700, respectively. In summary, in these four indices, Wuchang District and Jianghan District perform the best, indicating that the QHS in Wuchang District and Hanyang District in the central urban area of Wuhan is at a relatively high level, while there is room for improvement in all aspects of the QHS in Hongshan District and Qiaokou District.

3.4. A Probe into the Factors Affecting QHS

The dataset was partitioned into a training set (80%) and a test set (20%). The input layer consisted of 29 impact factor indicators from Table 2, with standardized and normalized score data serving as input. The output layer comprised the EHI, HCI, CTI, and QHSI indices. Subsequently, a model training was conducted using a hidden layer consisting of 10 neurons, with a Gaussian activation function, and MSE as the loss function (F) using learning factors α and β set at 0.1. To prevent overfitting, L2 regularization with a coefficient of 0.001 was added to the loss function to suppress excessive weights and enhance model generalization. Additionally, an early stopping strategy was employed during the iterative optimization process using gradient descent; the training process ceased when the validation set error did not decrease over five consecutive iterations. In the later stage of this study, the model architecture was expanded to include 10 to 30 neurons in the hidden layer. Subsequently, the MSE, R², and MAPE for the prediction effect of different configurations of neurons in the hidden layer on the ecological habitation ($A_1$), health and comfort ($A_2$), convenient transportation ($A_3$), and QHS index were calculated, resulting in 252 data points across four groups. The specific line graph is depicted in Figure 10. During each training process, the indicators on the validation set did not significantly worsen compared to the training set, indicating no significant overfitting of the model.

Upon analyzing the curve in Figure 10a, it is evident that the configuration with 28 neurons in the hidden layer generated the lowest MSE of approximately 0.0185, the highest R² value of approximately 0.8531, and the MAPE of only 15.76%. These findings indicate that the model with 28 neurons in the hidden layer represents the optimal configuration for predicting ecological habitation ($A_1$). Additionally, it was found that configurations with 22, 21, and 25 neurons in the hidden layer represent the optimal configurations for predicting health and comfort ($A_2$), convenient transportation ($A_3$), and the QHS index, respectively. A comparative analysis of the prediction effects of the models revealed that the optimal configuration for the QHS index model generated the lowest MSE and MAPE values, at 0.0123 and 12.68%, respectively, and also yielded the largest R² value of 0.9415. This validates the effectiveness and accuracy of our indicator selection.

To further validate these optimal configurations, MATLAB was used to conduct an importance analysis of the respective independent variable indicators for each model’s optimal configuration. The RBFNN stored the nonlinear mapping relationship between input and output signals through training. The connection weights between the input layer neurons and the hidden layer neurons, as well as the connection weights between the hidden layer neurons and the output layer neurons, were derived from the trained RBFNN, and the relative contributions of each influencing factor were calculated using the GARSON algorithm. The results are shown in Figure 11.

From Figure 11a, at a macro level, the indicator “pluralism and tolerance” ($B_2$) exhibits the highest importance for ecological habitability $\left(A_1\right)$, reaching 0.343, while “innovative vitality” ($B_4$) shows the lowest importance at only 0.186. On a micro level, the indicators “cultural facilities” ($B_{12}$), “preservation of historic districts” ($B_{13}$), “minimum subsistence level” ($B_{26}$), and “market environment” ($B_{43}$) all display an indicator importance for ecological habitation ($A_1$) that meets or exceeds 0.060, constituting Most High Importance Indicators (MHIIs). Therefore, the most immediate and effective approach to rapidly improve urban ecological habitation ($A_1$) index would be to take measures to enhance the levels of $B_{12}$, $B_{13}$, $B_{26}$, and $B_{43}$. This finding aligns with the research conclusion of Zhang et al. [68]. Conversely, “landmark building” ($B_{11}$), “property management” ($B_{32}$), “street hygiene” ($B_{33}$), “street lighting management” ($B_{36}$), “emergency treatment measures” ($B_{39}$), “talent introduction policy” ($B_{41}$), and “job opportunities” ($B_{42}$) demonstrate an indicator importance for ecological habitation ($A_1$) ranging between 0.040 and 0.060, qualifying as High Importance Indicators (HIIs). Indicators such as “historic buildings utilization” ($B_{14}$), “attraction to tourists” ($B_{15}$), “house price acceptability” ($B_{21}$), “rent acceptability” ($B_{22}$), “extent of regulation of the housing rental market” ($B_{23}$), “foreign population friendliness” ($B_{24}$), “curb ramp setup” ($B_{2A}$), and “youth appeal” ($B_{45}$) have an indicator importance for ecological habitation ($A_1$) below 0.020, classified as Low Importance Indicators (LIIs), indicating a lower correlation with ecological habitation ($A_1$). Other indicators have an indicator importance for ecological habitation ($A_1$) between 0.020 and 0.040, representing Intermediate Importance Indicators (IIIs).

The analysis of Figure 11b demonstrates that, from a macro perspective, the indicator “degree of tidiness” ($B_3$) holds the highest importance for health and comfort ($A_2$), reaching 0.365, while the lowest importance is attributed to “landscape features” ($B_1$) at merely 0.180. At the micro level, no indicators have an importance for health and comfort ($A_2$) reaching or exceeding 0.060, indicating that no single indicator significantly affects health and comfort ($A_2$). Conversely, indicators such as “landmark building” ($B_{11}$), “cultural facilities” ($B_{12}$), “preservation of historic districts” ($B_{13}$), “vulnerable group care” ($B_{25}$), “minimum subsistence level” ($B_{26}$), “Shantytown rehabilitation level” ($B_{28}$), “residential garbage classification” ($B_{31}$), “property management” ($B_{32}$), “street hygiene” ($B_{33}$), “pole management” ($B_{35}$), “street lamp management” ($B_{36}$), “emergency treatment measures” ($B_{39}$), “job opportunities” ($B_{42}$), and “market environment” ($B_{43}$) demonstrate a level of importance for ecological habitation ($A_1$) ranging between 0.040 and 0.060, qualifying as HIIs, almost twice the number of HII indicators for ecological habitation ($A_1$). This reflects the high comprehensive nature of health and comfort ($A_2$), which aligns with the conclusion drawn by Wang et al. [69].

The importance of indicators for convenient transportation ($A_3$) is shown in Figure 11c. At a macro level, the indicators with the highest and lowest importance for convenient transportation ($A_3$) are “degree of tidiness” ($B_3$) and “innovative vitality” ($B_4$), at 0.369 and 0.180, respectively. On a micro level, no indicators have an importance for convenient transportation ($A_3$) reaching or exceeding 0.060, with indicators such as “distinctive buildings” ($B_{11}$), “cultural facilities” ($B_{12}$), “preservation of historic districts” ($B_{13}$), “vulnerable group care” ($B_{25}$), “minimum subsistence level” ($B_{26}$), “Shantytown rehabilitation level” ($B_{28}$), “residential garbage classification” ($B_{31}$), “property management” ($B_{32}$), “street hygiene” ($B_{33}$), “pole management” ($B_{35}$), “street lamp management” ($B_{36}$), “emergency treatment measures” ($B_{39}$), “talent introduction policy” ($B_{41}$), “job opportunities” ($B_{42}$), and “market environment” ($B_{43}$) comprising 15 HII indicators, alongside 8 LII indicators, including “historic buildings utilization” ($B_{14}$), “attraction to tourists” ($B_{15}$), “house price acceptability” ($B_{21}$), “rent acceptability” ($B_{22}$), “extent of regulation of the housing rental market” ($B_{23}$), “foreign population friendliness” ($B_{24}$), “curb ramp setup” ($B_{2A}$), and “youth appeal” ($B_{44}$). This illustrates the high variance of the convenient transportation ($A_3$) indicator, which is in line with the conclusion of Bin et al. [70].

Figure 11d presents an analysis of the importance of various influencing indicator factors for the QHS. At a macro level, the indicator “degree of tidiness” ($B_3$) demonstrates an importance of 0.370 for the QHS, the highest among the secondary indicators; in contrast, the importance of “innovative vitality” ($B_4$) is only 0.155, lower than that of “unique characters” ($B_1$) at 0.196 and “diversity and inclusiveness” ($B_2$) at 0.282, making it the lowest of the four secondary indicators. At a micro level, “cultural facilities” ($B_{12}$) is the only indicator reaching or exceeding an importance of 0.060 for the QHS, due to its high importance for ecological habitation ($A_1$), indirectly influencing the QHS. Lee et al. [71] specifically studied the influence of different facilities on people’s satisfaction with urban environments. Additionally, among the indicators that influence the QHS, 14 of them, including “distinctive buildings” ($B_{11}$), “preservation of historic districts” ($B_{13}$), “vulnerable group care” ($B_{25}$), “minimum subsistence level” ($B_{26}$), “Shantytown rehabilitation level” ($B_{28}$), “residential garbage classification” ($B_{31}$), “property management” ($B_{32}$), “street hygiene” ($B_{33}$), “pole management” ($B_{35}$), “street lighting management” ($B_{36}$), “emergency treatment measures” ($B_{39}$), “talent introduction policy” ($B_{41}$), “job opportunities” ($B_{42}$), and “market environment” ($B_{43}$) are deemed HIIs, providing direction for improving the QHS. Furthermore, the LII, HII, and III indices for the QHS align more closely with those of convenient transportation ($A_3$), health and comfort ($A_2$), and ecological habitation ($A_1$), respectively. This suggests a high degree of variability in the indicators influencing the QHS.

4. Discussion

(1)

For the EHI, the importance of cultural facilities ($B_{12}$), preservation of historic districts ($B_{13}$), minimum subsistence level ($B_{26}$), and market environment ($B_{43}$) to ecological habitation ($A_1$) reached 0.069, 0.067, 0.061, and 0.060, respectively, demonstrating a significant influence on ecological habitation ($A_1$). These factors indirectly mitigate the impact of degree of tidiness ($B_3$) on ecological livability. The importance of various secondary indicators is relatively balanced [72]. In contrast, the importance of tertiary indicators displays significant differences. Therefore, it is crucial to improve the level of cultural facilities, enhance the minimum living protection level, strengthen the preservation of historic districts, and improve the market environment to elevate ecological livability. Furthermore, it is worth noting that, in August 2024, the inclusion of the Tahuayuan Historic District in Wuchang District, Wuhan City, in the “List of Replicable Experience Practices for the Protection and Utilization of Historic and Cultural Districts (First Batch)” highlights the emphasis placed by management on the preservation of historical districts.

(2)

HCI is a comprehensive index, as no indicator showing an importance exceeding 0.06 for community health and comfort ($A_2$). Even the least important indicator, curb ramp setup ($B_{2A}$), has an importance of 0.012. The range of importance for 29 indicators is 0.042, with a variance of $1.86\times {10}^{-4}$, both the lowest among the four indices in this study. Various characteristic landscapes in the city contribute to residents’ relaxation after work and study [73]. An inclusive urban environment makes residents feel the warmth of the city, while clean and tidy urban hygiene has a sustained positive impact on residents’ physical health [74,75]. Innovation and vitality can create new job opportunities for residents, all of which have important effects on residents’ health and comfort. Therefore, to improve the level of health and comfort requires concerted efforts in multiple aspects, particularly in enhancing the minimum living protection level and improving street hygiene management.

(3)

In the analysis of CTI, the importance of cultural facilities ($B_{12}$), street hygiene ($B_{33}$), preservation of historic districts ($B_{13}$), market environment ($B_{43}$), and minimum subsistence level ($B_{26}$) to convenient transportation ($A_3$) reached 0.055, 0.055, 0.054, 0.051, and 0.051, respectively. While the importance of these indicators did not reach 0.06, the difference between their maximum importance and the upper quartile is only $9.70\times {10}^{-3}$, with a variance as low as $4.20\times {10}^{-6}$, both of which are the lowest among the four indices in this study. Conversely, the variance between their minimum and upper quartile is $1.75\times {10}^{-4}$, the highest among the four indices. These characteristics reflect the concentration of high-importance indicators for convenient transportation and the dispersion of intermediate- and low-importance indicators [76]. To enhance the level of convenient transportation in a city, it is essential to prioritize the development of cultural facilities, street hygiene management, preservation of historic districts, improvement of the market environment, and ensuring minimum living standards. The calculation results demonstrate that market environment $\left(B_{43}\right)$ has an importance value of 0.051 for CTI. Additionally, it is noteworthy that, in September 2024, the National Fair Competition Conference opened in Wuhan, Hubei Province, showcasing the continuous efforts of Wuhan City in implementing fair competition policies and improving the favorable environment of the market.

(4)

Finally, for the QHSI, there is significant macro-level divergence. The importance variance of the dimensions of unique characters ($B_1$), diversity and inclusiveness ($B_2$), degree of tidiness ($B_3$), and innovative vitality ($B_4$) is $9.08\times {10}^{-3}$, higher than those of EHI, HCI, and CTI, reflecting a substantial difference in the importance of indicators at the macro level, with a range of 0.215, also the highest among the four indices. At the micro level, cultural facilities ($B_{12}$), preservation of historic districts ($B_{13}$), and street hygiene ($B_{33}$) exhibit an importance of 0.060, 0.057, and 0.056, presenting a strong correlation with the QHS. To improve the QHS, priority should be given to the development of cultural facilities, preservation of historic districts, and street hygiene, while paying attention to the construction of a neat and orderly urban environment. These findings are in line with the conclusion reached by Chen et al. [74]. For example, in the construction of the Optics Valley Cultural Center in Wuhan, efforts are being made to create a cultural complex with various cultural facilities to meet the needs of different age groups, aiming to achieve an age-friendly, open, and shared environment.

In summary, there is a certain degree of similarity in the importance indicators of these four indices. Preservation of historic districts ($B_{13}$) and minimum subsistence level ($B_{26}$) show importance exceeding 0.050 across the four indices, with average values reaching 0.057 and 0.055, respectively. Additionally, the average importance values of cultural facilities ($B_{12}$), street hygiene ($B_{33}$), and market environment ($B_{43}$) are all above 0.050, at 0.058, 0.054, and 0.051, respectively. These indicators all play a critical role in enhancing these index scores.

5. Conclusions and Future Prospects

5.1. Conclusions

As global urbanization and industrialization advance, accumulated urban issues are becoming increasingly prominent. These issues severely constrain the sustainable development of cities and the enhancement of the QHS. Many government managers and institutions are striving to establish QHS evaluation systems and investigate influencing factors to improve urban residents’ satisfaction. This paper proposes a method based on a combination of VIKOR, RBFNN, and GARSON to achieve feature extraction for the QHS and deeply understand the influencing mechanism of the QHS within the living environment. The main conclusions obtained are as follows.

(a)

Taking the QHSI of Wuhan City, China, as an example, the feasibility and effectiveness of the proposed method are validated. The traditional methods of QHS analysis exhibit limited robustness and comprehensiveness, making it difficult to comprehensively construct the QHS evaluation system and understand the relationships and mechanisms of various influencing factors. In contrast, the proposed combined approach adeptly explores the relationship between the influencing indicators and the QHS evaluation indicators. The results indicate that the overall QHSI of Wuhan City is 0.725, with EHI, HCI, and CTI being 0.759, 0.716, and 0.712, respectively.

(b)

Drawing on an extensive literature review and expert consultations, this paper quantifies QHS from 29 dimensions, which can be categorized into three aspects, and scrutinizes their impact on the QHS from 29 dimensions across four aspects. At the macro level, degree of tidiness ($B_3$) is identified as the key aspect influencing the QHS. At the micro level, cultural facilities ($B_{12}$), preservation of historic districts ($B_{13}$), and street hygiene ($B_{33}$) are identified as the primary influencing factors, exhibiting high importance and driving force in distinct aspects of the QHS. To improve a city’s QHS, it is essential to enhance the development of cultural facilities, protect historic districts, and improve street hygiene management.

(c)

Without a clear understanding of the intrinsic relationships between these influential factors and the construction of indicators, social managers will find it challenging to implement corresponding measures of varying importance levels for influential factors, potentially resulting in suboptimal outcomes. Compared to traditional analytical methods, the proposed combined method utilizes the VIKOR algorithm to assess the QHS, applies RBFNN for the nonlinear modeling of the 29 influencing factors, and finally employs the GARSON algorithm for importance analysis, revealing the significance of each indicator. This approach not only streamlines the efficiency of data processing and analysis but also demonstrates the robustness and generalization performance of the model, making the conclusions more stable and reliable.

5.2. Limitations and Future Work

Building on previous research on the QHS, this paper innovatively introduces a new method based on RBFNN-GARSON and proposes a multi-index evaluation system feature extraction importance model, providing a fresh perspective for the assessment and optimization of the QHS. However, our approach has some limitations that need to be addressed in future research.

(1)

The evaluation model of the QHS construction and the analysis of its influencing factors in this paper are static, while the evaluation method of the QHS and the formation of influencing factors are part of an ongoing and dynamic process. The types and status of the construction indicators and influencing factors of the QHS evolve continuously across different stages. In future studies, dynamic analysis methods such as the Hidden Markov Model (HMM), Vector Autoregression Model (VAR), and Structural Equation Model (SEM) can be leveraged for further analysis [77,78,79].

(2)

This paper utilizes a machine-learning-based mixed method to construct a model for identifying key indicators affecting the QHS. However, this model lacks interpretability and validation, hindering its ability to effectively explain how each indicator impacts the QHS. We can enhance the model’s accuracy and robustness by establishing a system dynamics model to simulate the dynamic changes in the QHS [80] or conducting cross-validation and hold-out validation [81,82] to validate the model’s accuracy and robustness.

(3)

The original data used in this study are relatively limited, mainly derived from personal preferences of residents, and the determination of weights relies on the subjective opinions of a few experts, thus carrying a certain degree of subjectivity. Future research could consider incorporating a wider range of expert opinions, as well as integrating multiple sources of data [12] such as statistical data from city yearbooks, public opinion data, sentiment analysis, etc. [83,84]. Additionally, macroeconomic data like urban unemployment rates could be integrated into the analysis of the QHS to provide a more comprehensive and objective evaluation of the construction of evaluation indicators and analysis of influencing factors.