Unraveling Spatial Nonstationary and Nonlinear Dynamics in Life Satisfaction: Integrating Geospatial Analysis of Community Built Environment and Resident Perception via MGWR, GBDT, and XGBoost

Abstract

Rapid urbanization has accelerated the transformation of community dynamics, highlighting the critical need to understand the interplay between subjective perceptions and objective built environments in shaping life satisfaction for sustainable urban development. Existing studies predominantly focus on linear relationships between isolated factors, neglecting spatial heterogeneity and nonlinear dynamics, which limits the ability to address localized urban challenges. This study addresses these gaps by utilizing multi-scale geographically weighted regression (MGWR) to assess the spatial nonstationarity of subject perceptions and built environment factors while employing gradient-boosting decision trees (GBDT) to capture their nonlinear relationships and incorporating eXtreme Gradient Boosting (XGBoost) to improve predictive accuracy. Using geospatial data (POIs, social media data) and survey responses in Suzhou, China, the findings reveal that (1) proximity to business facilities (β = 0.41) and educational resources (β = 0.32) strongly correlate with satisfaction, while landscape quality shows contradictory effects between central (β = 0.12) and peripheral zones (β = −0.09). (2) XGBoost further quantifies predictive disparities: subjective factors like property service satisfaction (R² = 0.64, MAPE = 3.72) outperform objective metrics (e.g., dining facilities, R² = 0.36), yet objective housing prices demonstrate greater stability (MAPE = 3.11 vs. subjective MAPE = 6.89). (3) Nonlinear thresholds are identified for household income and green space coverage (>15%, saturation effects). These findings expose critical mismatches—residents prioritize localized services over citywide economic metrics, while objective amenities like healthcare accessibility (threshold = 1 km) require spatial recalibration. By bridging spatial nonstationarity (MGWR) and nonlinearity (XGBoost), this study advances a dual-path framework for adaptive urban governance, the community-level prioritization of high-impact subjective factors (e.g., service quality), and data-driven spatial planning informed by nonlinear thresholds (e.g., facility density). The results offer actionable pathways to align smart urban development with socio-spatial equity, emphasizing the need for hyperlocal, perception-sensitive regeneration strategies.

1. Introduction

As rapid urbanization reshapes global communities, coupled with profound socioeconomic transformations, it has exposed communities to multifaceted challenges—economic decline, aging infrastructure, weakened social cohesion, and diminished cultural vitality [1,2,3]. The 2023 SDGs report calls for making cities inclusive, safe, resilient, and sustainable, highlighting the urgency of addressing these issues through regeneration actions and equitable development at the community level [4,5]. Aligning with this global call, China’s “14th Five-Year Plan” prioritizes urban regeneration strategies that directly target the improvement in community spatial quality and the well-being of residents [6,7]. In this context, urban regeneration policies are not merely necessary but critical for enhancing the quality of life (QoL) through efficient public services, transportation networks, and green spaces [8,9,10]. Life satisfaction, an essential metric for evaluating community QoL, is becoming a focal point in urban regeneration efforts [11].

However, the underlying drivers remain a subject of ongoing debate: whether and to what extent life satisfaction is primarily shaped by the built environment or if subjective perception holds greater sway. Moreover, the complexity of these drivers still defies a unified academic measure; current research has yet to fully unravel the interplay between subjective perceptions and objective environmental factors, particularly when accounting for spatial heterogeneity and nonlinear effects [12,13]. Existing research has polarized into two camps: one focusing on the objective material space, such as housing conditions, infrastructure, and public space design [14,15], and the other on subjective aspects, using surveys to assess psychological states, social relationships, and personal fulfillment across different demographics [16]. Furthermore, the interplay between subjective perceptions and built environment factors, especially when considering spatial heterogeneity and nonlinear effects at different geographical scales, remains underexplored [17,18,19].

The rapid development of big data and spatial econometrics has opened new avenues for examining how community environments impact life satisfaction. Tools like geographic information systems (GIS) and remote sensing technologies now allow for comprehensive spatial analysis, enabling the construction of evaluation frameworks that incorporate points of interest (POI), land use, and built environment characteristics [20,21,22]. Existing urban regeneration policies often rely on qualitative judgments and lack the quantitative, spatially informed support needed for precision. Identifying key factors that affect residents’ life satisfaction through rigorous analysis—and translating these insights into actionable urban regeneration strategies—remains a pressing challenge [23]. Moreover, the multi-scale geographically weighted regression (MGWR) method has proven effective in capturing the spatial variability of life satisfaction determinants [24]. Simultaneously, machine learning models such as gradient-boosting decision trees (GBDT) offer a robust approach to uncovering nonlinear interactions across various dimensions [25]. These methods clarify the complex dynamics shaping life satisfaction by enabling precise mapping of environmental quality and facility distribution, offering actionable insights for urban regeneration policies [26].

Our study directly addresses a critical gap in current urban QoL research. Existing studies often fail to bridge the gap between residents’ subjective perceptions and objective environmental measurements, resulting in policy interventions that lack effectiveness. To solve this, we develop a dual-dimensional framework that fuses subjective perception data with built environment geospatial factors. This approach is designed to overcome spatial misalignments and mechanistic disconnects observed in prior research. We employ MGWR to assess how these elements interact across diverse spatial contexts, capturing the nonstationarity of life satisfaction influences. MGWR allows us to pinpoint regional variability in the factors that drive life satisfaction. Complementing this, our GBDT model uncovers nonlinear relationships between individual perceptions and built environment factors at the urban scale, offering a nuanced view of these dynamics. Furthermore, to boost our model’s predictive power, we integrate eXtreme Gradient Boosting (XGBoost), a state-of-the-art machine learning algorithm that refines accuracy by analyzing the interplay of subjective and objective factors. By combining MGWR, GBDT, and XGBoost, our study delivers a comprehensive and robust model that accounts for both spatial variability and the complex dynamics linking subjective perceptions to built environment factors. This integrated approach enables us to answer three pressing questions:

• What are the key factors of consistency and heterogeneity between subjective perceptions and objective space in shaping life satisfaction?
• How do spatially measured factors exhibit significant nonlinear relationships with life satisfaction?
• How can targeted urban regeneration strategies be formulated based on a comprehensive understanding of life satisfaction drivers?

In sum, our research not only fills a crucial void by reconciling subjective and objective analyses but also offers actionable insights for more precisely targeted urban regeneration policies. The structure of this paper is as follows (see Figure 1): Section 2 reviews the literature on subjective perception and built environment factors affecting life satisfaction and identifies current research gaps. Section 3 details the research design, data sources, and case study methodology. Section 4 presents the MGWR and GBDT analysis results, emphasizing how subjective and objective spatial conditions interact to shape life satisfaction. Section 5 constructs an integrated predictive model based on XGBoost, systematically examines the policy implications for urban regeneration, and offers concrete policy recommendations grounded in empirical data. Finally, Section 6 summarizes the key findings and outlines directions for future research.

2. Literature Review on Life Satisfaction Influential Factors

2.1. Subjective Perception and Physical Environment Factors Influence Life Satisfaction

Life satisfaction is defined as an individual’s overall cognitive evaluation of their QoL based on self-determined criteria, forming a core dimension of subjective well-being [27]. Existing studies highlight that life satisfaction emerges from the interplay between subjective perceptions and objective environmental conditions [28,29]. Subjective perception encompasses psychological well-being, the quality of social relationships, and a sense of personal achievement, while the physical environment refers to objective conditions such as housing quality, infrastructure, and service facilities [30,31]. The interplay between these factors, though widely recognized, remains contested in terms of their relative importance.

Sociologists such as David Harvey and Saskia Sassen emphasize the profound role of individual perception [32,33], shaped by psychological and social factors, in determining residents’ life satisfaction. Their research suggests that subjective perceptions often outweigh objective environmental conditions in shaping well-being [28,29]. For instance, a study found that residents’ subjective perceptions of air quality have a more significant negative impact on their happiness than actual air quality data [34]. Furthermore, residents’ satisfaction with green spaces is closely linked to mental health, particularly among vulnerable populations, whereas actual greenery conditions may not correlate significantly with health outcomes [35]. The interaction between perceived pollution risks and green exposure notably affects subjective well-being, indicating that both improving green spaces and reducing pollution should be prioritized to enhance QoL [36]. Research on migrants further underscores that subjective well-being is influenced by social relationships and the living environment, with factors like relative deprivation, social support, and neighborhood conditions playing a significant role in life satisfaction [37]. This evidenced that personal interpretations of one’s environment play a critical role in shaping overall well-being, even when objective conditions remain constant.

Nevertheless, urban planners and environmental behaviorists, including Kevin Lynch and Jane Jacobs, highlight the critical importance of the physical environment in influencing behavior and life satisfaction. Lynch’s work suggests that urban spatial layout directly impacts emotional and cognitive experiences [38], while Jacobs argues that vibrant, well-designed public spaces promote social connections and enhance life satisfaction [39]. Empirical studies substantiate this by showing that factors such as increased green space correlate with higher reported life satisfaction [40]. Additionally, research by Xie et al. highlights that the accessibility of the physical environment significantly affects the health of the elderly [41]. Moreover, research from Oslo demonstrates that commuting environments significantly affect neighborhood and job satisfaction, influencing overall well-being [42]. This evidence reinforces the notion that physical environments can shape subjective experiences and thus contribute to life satisfaction [30,31].

Despite these extensive studies, the debate over the relative importance of subjective perception versus physical environment remains unresolved. Scholars argue for an integrated understanding, suggesting that life satisfaction is best explained through the dynamic interaction between subjective and objective conditions [43]. Still, factors jointly shape well-being through complex interactions [44]. For example, improved physical conditions may not enhance satisfaction if expectations rise in tandem [45]. Moreover, the social comparison theory posits that individuals’ subjective well-being is influenced by changes in their surroundings, particularly when community-wide improvements reduce feelings of relative deprivation [46]. Thus, the relationship between subjective perception and the physical environment is neither linear nor one-dimensional but multi-faceted and interactive, requiring a deeper understanding of residents’ perceptions and reactions to environmental changes when assessing life satisfaction.

2.2. Nonlinear Factors Affecting Life Satisfaction

Examining life satisfaction demands a nuanced understanding of the nonlinear relationships between various influencing factors [47]. Recent studies have identified key elements such as income, age, and mental health, which exhibit nonlinear effects on well-being [48,49,50]. For instance, income has a diminishing marginal effect on life satisfaction, meaning that after a certain threshold, increases in wealth bring progressively smaller gains in happiness [51]. Similarly, the initial increase in green space significantly improves well-being, but beyond a certain point, additional greenery yields diminishing returns [52]. These findings suggest that the relationship between life satisfaction and its contributing factors is far more complex than linear models suggest.

Housing conditions, a critical component of the built environment, also exhibit nonlinear effects on life satisfaction. Studies show that improvements in living environments initially lead to marked increases in satisfaction, yet these effects gradually diminish over time [53,54]. This aligns with Maslow’s hierarchy of needs, which posits that once basic housing needs are met, individuals’ focus shifts to higher-order needs, such as social belonging and self-actualization [55]. The relationship between housing factors like space and cost similarly reflects diminishing returns, suggesting that while larger living spaces or higher housing prices may initially boost satisfaction, beyond a certain point, these benefits taper off.

Community facilities, commuting conditions, and social networks underscore the nonlinear dynamics influencing life satisfaction [56,57]. Commuting time, for example, is a well-documented determinant of life satisfaction, with shorter, less stressful commutes associated with significant improvements in well-being [58,59]. Interestingly, car ownership is linked to higher satisfaction compared to public transportation users, further illustrating the intricate connections between personal resources and subjective well-being [60]. Such findings emphasize the complex, interwoven nature of individual, social, and environmental factors in determining life satisfaction, underscoring the need for comprehensive and multifaceted approaches to understanding well-being [61].

2.3. Methodological and Assessment Gaps

The nonlinear and spatially nonstationary relationships between life satisfaction and its determinants remain underexplored. While considerable research has focused on subjective perceptions (e.g., mental health and personal achievement) and objective physical environments (e.g., housing conditions and socio-economic status), a comprehensive comparison of the relative importance of these factors is lacking. Most studies isolate subjective or objective influences, neglecting how they interact across different contexts. Furthermore, the interplay between these two dimensions and how this interaction shapes life satisfaction remains poorly understood. This gap limits our ability to grasp the mechanisms behind life satisfaction in complex, real-world scenarios [62].

The spatial heterogeneity of life satisfaction at finer urban scales also requires deeper examination [63]. Although some studies have recognized the spatial nonstationarity of life satisfaction, few have analyzed its micro-medium scale variability within cities. Differences in life satisfaction across various geographic regions and community environments, particularly how local environmental conditions shape these differences, have been insufficiently addressed. The spatial disparities in life satisfaction reflect the broader limitations in handling spatial heterogeneity in current research [64].

Methodologically, existing research oversimplifies the dynamic interplay among multiple determinants influencing life satisfaction [65]. Most investigations concentrate on single variables or a narrow set of factors, failing to capture the complex interactions that influence life satisfaction [66]. For instance, some investigations employ SEM or PLS-SEM to study satisfaction [67], while others leverage 5D theory with AHP and spatial analysis to probe urban vitality [68,69] or apply MGWR to assess spatial heterogeneity [70] and even utilize GIS and image segmentation for environmental evaluation and enhancement [71,72,73,74]. Even when studies attempt integration, they typically rely on qualitative assessments and lack robust quantitative frameworks, thus missing critical nonlinear dynamics [75]. There is an urgent need for innovative tools and methods that systematically evaluate how multiple factors interact to affect life satisfaction, such as advanced algorithm-driven nonlinear techniques like GBDT and predictive tools like XGBoost [76].

2.4. Research Framework

This study proposes an integrated analytical framework to unravel the spatial nonstationarity and nonlinear dynamics of life satisfaction among Suzhou residents by synthesizing geospatial analytics, machine learning, and multi-source data (Figure 2). The framework tackles spatial heterogeneity and nonlinear interactions between urban built environments and resident perceptions through a hybrid approach that combines GIS, MGWR, GBDT, and XGBoost. Below, we detail the structure, functional roles, advantages, limitations, and innovations of our framework.

Structure Description: Our framework integrates subjective survey data (“small data”) with objective geographic “big data” (e.g., POIs, geotagged data like Weibo check-ins from social media platforms) and social demographics, thereby capturing the multidimensional facets of urban conditions. Methodologically, it visualizes spatial distributions of life satisfaction via graded mapping using GIS and quantifies spatially varying relationships between built environment factors and life satisfaction with MGWR. It then identifies nonlinear interactions among subjective factors—such as social interaction and commuting stress—using GBDT and further refines predictive accuracy through XGBoost, which synthesizes nonlinear patterns from GBDT and spatial effects from MGWR.

Functional Roles: GIS serves as the geospatial backbone, enabling data visualization and the identification of spatial patterns. MGWR addresses spatial nonstationarity by estimating location-specific coefficients for built environment variables (e.g., POI density, commuting distance). GBDT uncovers nonlinear thresholds and interactions in subjective factors (e.g., age–income interplay, social interaction intensity), while XGBoost enhances predictive performance by integrating outputs from both MGWR and GBDT, thereby reducing overfitting.

Advantages and Limitations: MGWR resolves spatial heterogeneity and GBDT analytics could capture nonlinearity characteristics combined with the prediction model via XGBoost, thus overcoming the oversimplification inherent in conventional models. Our approach robustly integrates qualitative perceptions (via questionnaire surveys) with quantitative big data (POIs, social media), yielding spatially explicit coefficients and optimized prediction accuracy. Nevertheless, the framework entails computational complexity: MGWR demands significant processing power for bandwidth optimization and GBDT/XGBoost requires careful hyperparameter tuning.

Innovation of the Framework: This study is the first to synergize MGWR (capturing spatial nonstationarity) with GBDT (addressing nonlinearity) and XGBoost (a machine learning-based prediction model) in the analysis of objective and subjective influential factors of life satisfaction. It bridges macro-level spatial patterns (via GIS/MGWR) with micro-level perceptual dynamics (via GBDT/XGBoost) and yields policy-relevant insights by identifying spatially targeted interventions (e.g., optimizing POI layouts in low-satisfaction zones) and nonlinear thresholds (e.g., critical commuting time limits).

To summarize, initially, we used the OLS model as a benchmark for estimation [77]. Then, GWR models were adopted and tested to reveal underlying processes and address spatial heterogeneity within datasets [78]. Compared to OLS, GWR improves model fitting by capturing spatial variations at multiple scales. MGWR extends GWR by allowing each predictor to adapt its spatial scale, which is crucial for understanding complex spatial heterogeneity [79]. GBDT, unlike traditional regression models, does not rely on assumptions about variable distribution or causal relationships [80]. It effectively captures non-linearities and provides insights into the relative importance of predictors, making it increasingly valuable in urban studies linking subjective perceptions and life satisfaction. XGBoost, an advanced version of GBDT, helps mitigate overfitting and enhances model performance [81,82].

3. Data and Methods

3.1. Research Area

Suzhou, a prefecture-level city in the core of the Yangtze River Delta, is renowned for its livability, robust economy, and iconic classical Chinese gardens [83]. As a pioneer in China’s economic transformation, Suzhou rapidly industrialized post-reform, driving its population to 12 million [84]. However, recent years have seen the city facing mounting challenges, including economic restructuring, urban–rural disparities, resource constraints, and environmental pressures. A striking 40% of its population comprises migrants, adding complexity to the social fabric and offering a unique opportunity to integrate human-scale elements with mid-scale urban analysis [85].

Figure 3 illustrates the distribution of questionnaire sample points across Suzhou, highlighting the various community boundaries within the city. This map, which includes urban districts like Gusu, Suzhou Industry Park (SIP), Suzhou New District (SND), Xiangcheng, Wujiang, and Wuzhong, offers a visual representation of the distribution of residents’ life satisfaction samples. The green areas represent community boundaries, while the red sample points correspond to survey responses from residents in those communities. This dynamic urban landscape makes Suzhou an ideal case for examining the factors influencing life satisfaction, particularly given its complex socio-spatial environment [86,87].

3.2. Methods

3.2.1. MGWR Analysis

MGWR discusses the spatial relationship between the dependent variable/response variable and the independent variable/explanatory variable [88]. This method allows the analysis of changes in geographical phenomena at different spatial scales. MGWR relaxes the assumption that all simulation processes occur at the same spatial scale [89]. Traditional models need to perform initial OLS regression, which filters the data factors, and then perform MGWR analysis on this basis.

(1) 

OLS regression

In this study, OLS regression is used as one of the core methods to analyze the factors influencing life satisfaction among residents of Suzhou, where each observation of residents’ life satisfaction, $y_i$, can be described by the following basic formula:

$$y_i={\beta }_0+{\beta }_1{x}_1+{\beta }_2{x}_2+{\beta }_i{x}_i+\epsilon$$

(1)

$y_i$ is the residents’ life satisfaction index, representing the dependent variable; ${\beta }_0$ is the intercept; $x_1,x_2$, … $x_i$ stands for the independent variable referred to in the research framework (Figure 2); ${\beta }_0,{\beta }_1,\dots {\beta }_i$ is the regression coefficient; and ε is the error term, representing the difference between the model’s prediction for the ith observed value $y_i$ and the actual value.

(2) 

Spatial variation test (Moran’s l test)

In this study, we use Moran’s Index to evaluate spatial autocorrelation in a multivariate geographically weighted regression model to measure the overall spatial autocorrelation. When Moran’s l is greater than 0, the data has a positive spatial correlation, and the larger the value, the more obvious the spatial correlation. In this study, the spatial weights are based on k nearest neighbors. The global Moran’s Index is summarized below:

$$I={\displaystyle \frac{n}{W}}{\displaystyle \frac{{\Sigma }_{i=1}^N{w}_{ij}{\Sigma }_{j=1}^N\left(x_i-\overline{x}\right)\left(x_j-\overline{x}\right)}{{\Sigma }_{i=1}^N{\left(x_i-\overline{x}\right)}^2}}$$

(2)

The global Moran’s Index is denoted as $I$, where n represents the number of observations corresponding to the communities selected in Suzhou. $x_i$ and $x_j$ represent the observed values for the $i$-th and $j$-th communities in Suzhou, and $\overline{x}$ represents the observed values for communities selected in Suzhou. $w_{ij}$ signifies the spatial weight between the observations of the $i$-th and $j$-th communities, and $W$ is the sum of all spatial weight vertical $w_{ij}$.

(3) 

MGWR analysis

Using the MGWR model, the study assessed the correlation between subjective perception and objective space at each geographical location using spatial weighting techniques. The model can reflect the differences and connections between different regions. The MGWR model is summarized as follows:

$$y_i={\beta }_{b{w}_0}\left(u_i,u_i\right)+{\Sigma }_k{\beta }_{b{w}_k}{x}_{ik}+{\epsilon }_i$$

(3)

where $y_i$ represents the residents’ life satisfaction index, ${\beta }_{b{w}_0}\left(u_i,u_i\right)$ and ${\beta }_{b{w}_k}$ represent the intercept term at i and the regression coefficient corresponding to the independent subjective and objective variable $x_{ik}$ (a combination of subjective satisfaction ratings (from the questionnaire survey) and objective built environment data (including characteristics of the community and surrounding areas) listed in the research framework in Figure 2), which is the k-th variable at the geospatial position $i$, respectively. $\left(u_i,u_i\right)$ is the coordinate of the position $i$.

3.2.2. GBDT Analysis

In this study, the GBDT algorithm is used to explore the impact weights of various factors affecting social perception, with a particular focus on understanding the relative importance of each factor in influencing life satisfaction. GBDT is an ensemble learning method that builds a series of decision trees sequentially, where each tree corrects the errors made by the previous one. This approach is particularly effective in capturing complex non-linear relationships between the independent variables (such as social demographics, built environment, and commuting conditions) and the dependent variable (life satisfaction). $\widehat{f}\left(x\right)$ is adopted to explore the impact weights of a single impact factor at different levels. The specific formula is as follows:

$${\mathrm{r}}_{\begin{array}{c}\mathrm{mi}\end{array}}={\left[{\displaystyle \frac{\partial \mathrm{L}\left(\mathrm{y},\mathrm{f}\left({\mathrm{x}}_{\mathrm{i}}\right)\right)}{\partial \mathrm{f}\left({\mathrm{x}}_{\mathrm{i}}\right)}}\right]}_{\mathrm{f}\left(\mathrm{x}\right)-{\mathrm{f}}_{\mathrm{m}-1}\left(\mathrm{x}\right)},{\theta }_{{\mathrm{m}}_{\mathrm{j}}}=\mathrm{a}\mathrm{r}{\mathrm{g}}_{\theta }\min {\sum }_{{\mathrm{x}}_{\mathrm{i}}\in {\mathrm{R}}_{{\mathrm{m}}_{\mathrm{j}}}}\mathrm{L}\left({\mathrm{y}}_{\mathrm{i}}{\mathrm{f}}_{\mathrm{m}-1}\left(\mathrm{x}\right)+\theta \right)$$

(4)

$$f_m\left(x\right)=f_{m-1}\left(x\right)+{\sum }_{j=1}^J{\theta }_{mj}\left(x\in r_{mj}\right)$$

(5)

$$\widehat{f}=f_m\left(x\right)={\sum }_{m=1}^M·{\sum }_{j=1}^J\theta m_{j}I\left(x\in r_{m_i}\right)$$

(6)

where N represents the samples of subjective questionnaires, M represents the number of regression trees, J represents the number of regression tree leaf nodes, $I\left(x\right)$ represents the indicator function used to judge the elements in the set, and $\theta$ represents the transcendental parameter.

3.2.3. XGBoost Analysis

In this study, we employ the XGBoost algorithm for data modeling and predictive analysis. XGBoost is an ensemble method based on the gradient boosting framework, known for its high computational efficiency and robust predictive power, which makes it highly suitable for regression and classification tasks. Specifically, the objective function in XGBoost, as shown in Equation (7), is designed to minimize the discrepancy between the predicted value $\widehat{{\mathrm{y}}_i}$ and the actual value ${\mathrm{y}}_i$ (e.g., using squared error). In this case, ${\mathrm{y}}_i$ represents the subjective life satisfaction scores obtained from residents. Here, T represents the number of leaf nodes in the tree, while γ and ∂ are regularization coefficients to prevent overfitting. Equation (8) outlines the update process for the predicted values, where $f_t\left({\mathrm{x}}_i\right)$ refers to the regression tree built in the t-th iteration. The splitting optimization for each node is detailed in Equation (9), where the optimal node split is determined by maximizing the gain based on the first- and second-order gradient statistics for the left (G_L, H_L) and right (G_R, H_R) child nodes.

$$\mathrm{O}\mathrm{b}\mathrm{j}\left(\Theta \right)=\sum _{i=1}^{n}l\left({\mathrm{y}}_i,\widehat{{\mathrm{y}}_i}\right)+\sum _{k=1}^K[\gamma T+{\displaystyle \frac{1}{2}}\partial \sum _{j=1}^T{w}_j^2]$$

(7)

$${\widehat{{\mathrm{y}}_i}}^{\left(t\right)}={\widehat{{\mathrm{y}}_i}}^{(t-1)}+f_t\left({\mathrm{x}}_i\right)$$

(8)

$$Gain={\displaystyle \frac{1}{2}}\left[{\displaystyle \frac{{\mathrm{G}}_{\mathrm{L}}^2}{{\mathrm{H}}_{\mathrm{L}}+\partial }}+{\displaystyle \frac{{\mathrm{G}}_{\mathrm{R}}^2}{{\mathrm{H}}_{\mathrm{R}}+\partial }}-{\displaystyle \frac{{(\mathrm{G}}_{\mathrm{L}}^2+{\mathrm{G}}_{\mathrm{R}}^2)}{{\mathrm{H}}_{\mathrm{L}}+{\mathrm{H}}_{\mathrm{R}}+\partial }}\right]-\gamma$$

(9)

In this study, XGBoost is used to perform predictive analysis on significant subjective and objective factors affecting life satisfaction. Specifically, we utilize XGBoost to model and predict the relationships between these factors—such as income, housing conditions, and accessibility to various facilities—and life satisfaction. By applying XGBoost, we can identify which factors have the most substantial impact on life satisfaction while also uncovering the complex, nonlinear relationships between them. This predictive approach allows for more precise estimations and helps reveal threshold effects, providing deeper insights into how different factors interact and contribute to the overall well-being of residents.

3.3. Data Collection

The data collection for this study followed a structured, multi-dimensional approach, integrating subjective perception surveys with objective geospatial data. The subjective survey comprises four key components: social demographics, social interaction, the built environment, and the commuting environment (

Table 1. Details of subjective questionnaire and objective geospatial variables.

Items	Subjective Questionnaire Variable	Objective Geospatial Variable
Social Demographics	Housing price/Household Income/Household Registration Type/Age/Residence Duration/Educational Background/Family Members/Gender	Housing Price/Plot Ratio/Family Members/Construction Time
Social Interaction	Health Condition/Exercise Frequency/Neighborhood Relationship/Job Stability/Working Condition/Communication Frequency/Psychological Status/Financial Condition/Emotional Stability/Emotional State/Time Arrangement/Financial Stability	Weibo Check-in/Entertainment Check-in/Catering Check-in/Travelling Check-in/Facilities Check-in
Built Environment	Landscape Quality/Property Service/Business Facilities Convenience/Educational Facilities Accessibility/Housing Type/Living Units’ Area/Dwelling Height	Catering Facilities/Scenic Spots/Shopping Facilities/Public Facilities/Science and Education Cultural Facilities/Living Service Facilities/Leisure Facilities/Medical Service/Green Space Ratio
Commuting Environment	Transport Facilities/Accessibility Services/Environmental Safety/Mode of Transportation/Traffic Condition/Transport Greenery/Non-motorized Commuting Time/Motorized Commuting Time/Commuting Distance/Commuting Cost/Commuting Time/Environmental Aesthetic/Environmental Cleanliness	Transport Facilities/Road Service Facilities

). Social demographics cover variables such as housing price, household income, household registration, age, residence duration, education, family size, and gender. Social interaction factors include health status, exercise frequency, neighborhood ties, job stability, and working conditions. The built environment assesses landscape quality, property services, business and educational facility accessibility, housing type, unit area, and dwelling height. Commuting environment variables encompass transport infrastructure, traffic conditions, green transport routes, motorized and non-motorized commute times, distance, costs, environmental aesthetics, and cleanliness.

All subjective perception variables were measured using a 5-point Likert scale, where respondents rated their level of agreement or satisfaction on a scale ranging from “Strongly dissatisfied” to “Strongly satisfied”. This scale was applied to capture residents’ subjective satisfaction or perceived quality for each factor, ensuring consistency and comparability across different variables (Appendix A).

Objective geospatial data mirrored these categories. Social demographics included metrics like housing price, plot ratio, family size, and construction date. Social interaction was captured via social media check-ins (Weibo), covering entertainment, dining, travel, and public facilities. The built environment featured data on catering services, scenic areas, shopping, public facilities, educational and cultural services, medical care, and green space ratios. The commuting environment section analyzed transport and road service infrastructure.

The research team conducted surveys across key districts of Suzhou from 5 March to 25 March 2024, including Gusu, SIP, and SND. Using random sampling, the team collected 1535 responses, of which 1504 were valid, yielding a 98% response rate.

Table 2. Basic information of the survey samples (N = 1504).

Statistical Variable	Category	Sample Size	Percentage
Gender	Male	809	53.8%
	Female	695	46.2%
Age	20~35	1257	83.6%
	35~50	247	16.4%
Household Income	≤10,000 RMB	88	5.9%
	10,000–15,000 RMB	736	48.9%
	15,001–25,000 RMB	432	28.7%
	25,001–45,000 RMB	153	10.2%
	≥45,001 RMB	95	6.3%
Household Registration Type	Local household registration	815	54.2%
	Rural household registration	236	15.7%
	Urban floating population	205	13.6%
	Rural floating population	195	13.0%
	Collective household registration	53	3.5%

outlines the demographic breakdown of the respondents.

The study’s dataset, detailed in

Table 3. A detailed description of the geospatial and questionnaire data source.

Item	Description and Source	Quantity	Time
Questionnaire data	Personal attribute/community environment perception/social interaction and perception/commuter environment perception	1504 copies	2024
Basic community information	Housing price/construction time/greening rate/the community floor area ratio	2311 pieces	2023
Social media data	Weibo check-in data with text and geo-location. Accessed from: https://weibo.com. Accessed on 10 September 2024.	258,631 pieces	2023
POI data	Shopping/healthcare/tourist attractions/catering	319,597 polygons	2024

, includes data categories, sources, sample size, and collection periods. These comprehensive data points form the basis for analyzing the complex socio-spatial factors influencing life satisfaction across diverse urban settings in Suzhou. While the survey data used in this study were collected in 2024, certain spatial-related data, such as POI information, were updated to reflect 2024 data. However, due to limitations in data availability at the time of writing (2024), price data for communities and social media data from Weibo were only available for 2023. This discrepancy in data sources and timeframes has been duly noted and does not impact the overall analysis, as the temporal variation is minimal for the study’s objectives.

4. Results

4.1. MGWR Results Based on Subjective Perception Data

4.1.1. Spatial Correlation Test and OLS Regression Results Based on Subjective Data

To ensure the validity of spatial patterns and assess the suitability of model assumptions, the study applied Moran’s I and OLS tests before conducting MGWR. This combination strengthens the accuracy of subsequent spatial regression models, ensuring that the chosen methods properly account for the spatial dependencies in the data (

Table 4. Global Moran Index based on subjective questionnaire data.

Dependent Variable	Moran’s Index	Z Value	p Value	E(I)
Life satisfaction	0.07885	9.1398	0.0000	−0.0006

Notes: Z value is the standard deviation difference between the regression coefficient and zero. p value is used to evaluate whether the regression coefficient is significantly different from zero. E(I) represents spatial autocorrelation.

). A clear spatial clustering effect emerges for residents’ life satisfaction, with a significant positive Moran’s I value well above zero. This indicates a positive spatial correlation, meaning that communities with higher satisfaction tend to cluster. The significance level of 0.000 (<0.05) confirms the robustness of this relationship at the 5% significance level.

Table 5. OLS results with model variables.

Variable	Coefficient a	StdError	t-Statistic	Probability b	Robust_SE	Robust_Pr b	VIF c
Intercept	0.9896	0.2284	4.3371	0.000 ***	0.2304	0.0000 ***	N/A
Household registration type	−0.0554	0.0193	−2.8553	0.0043 *	0.0209	0.0081 *	1.1333
Household income	0.0389	0.0240	1.5573	0.1195 *	0.0287	0.1871 *	1.1055
Residence duration	0.1022	0.0345	3.0077	0.0026 **	0.0428	0.0171 *	1.1047
Housing type	−0.0694	0.0207	−3.3750	0.0007 **	0.0217	0.0013 **	1.0783
Housing price	0.0804	0.0303	2.6162	0.0089 *	0.0385	0.0372 *	1.4338
Property service	0.0790	0.0263	2.9961	0.0027 *	0.0315	0.0121 *	1.3656
Landscape quality	0.1090	0.0287	3.7943	0.0001 ***	0.0345	0.0016 **	1.3810
Business facilities	0.0945	0.0293	3.2197	0.0013 **	0.0337	0.0051 *	1.3931
Educational facilities accessibility	0.0616	0.0286	2.1510	0.0316 *	0.0355	0.0829 *	1.3244
Neighborhood relationship	0.0253	0.0156	1.6153	0.1065 *	0.0151	0.0943 *	1.0082
Financial condition	−0.0370	0.0159	−2.3190	0.0205 *	0.0151	0.0194 *	1.0076
Public facilities	0.1099	0.0272	4.0390	0.000 ***	0.0298	0.0001 ***	2.0455
Environmental safety	0.1047	0.0275	3.7992	0.000 ***	0.0344	0.0022 **	1.2671
Environmental aesthetic	−0.0280	0.0159	−1.7547	0.0795 *	0.0162	0.0797 *	1.0096
Transportation mode	0.0796	0.0283	2.8138	0.0049 *	0.0334	0.0172 *	1.2807
Motorized commuting time	−0.1255	0.0269	−4.6524	0.000 ***	0.0361	0.0005 ***	1.4579
Transport greenery	0.0432	0.0307	1.4087	0.1591 *	0.0358	0.2272 *	1.4623

*** p < 0.01, ** p < 0.05, * p < 0.1. Coefficient ^a: indicates the influence direction (positive or negative). Probability ^b: the p-value corresponding to coefficient a. Robust_Pr ^b: the p-value of coefficient a calculated considering heteroskedasticity and autocorrelation. VIF ^c**: an indicator test for the existence of multicollinearity among independent variables.

presents the results of the OLS regression, identifying 11 key variables for further analysis in the MGWR model. The findings reveal that housing type, financial condition, and motorized commuting time significantly reduce life satisfaction. Increased vehicle ownership and prolonged commutes worsen traffic congestion and commuting experiences, lowering overall well-being. Policies that expand public transportation and shorten commuting times could mitigate these negative effects. Conversely, household income, housing price, property service quality, business and public facilities, environmental safety, and transportation mode exhibit a positive correlation with life satisfaction. Higher income and rising property values often signal improved living standards, fostering greater satisfaction. Well-managed properties, accessible business facilities, and enhanced public infrastructure meet higher-order urban needs, reinforcing overall well-being. By leveraging MGWR, this study overcomes the spatial homogeneity constraints of OLS, capturing localized variations in life satisfaction determinants. The model quantifies nonlinear thresholds, offering a spatially explicit decision-making basis for targeted urban regeneration and policy interventions.

4.1.2. MGWR Analysis Based on Subjective Perception Data

After conducting the global Moran’s I test and OLS regression analysis, we proceeded with MGWR based on subjective perception data. The results revealed significant spatial heterogeneity across personal attributes, built environment, and commuting conditions with life satisfaction. Regarding the relationship between personal attributes, the MGWR analysis retained household registration type, residence duration, and financial condition as significant factors influencing life satisfaction. The color scale in MGWR figures has been defined as follows: green represents negative correlations, with deeper shades of green corresponding to significant negative values in the region. Red represents positive values, with deeper shades of red signifying significant positive correlations.

As shown in Figure 4(A1), Household registration exhibited slight spatial variation, generally showing a negative spatial correlation with life satisfaction. Rural and migrant households tend to report lower satisfaction. However, areas in the northern inner ring, southern Xiangcheng District, and parts of the industrial park displayed positive correlations, reflecting the aggregation of migrant populations and their spatial heterogeneity. Figure 4(A2) indicates that residence duration positively affects life satisfaction across most regions, except in industrial zones, where long-term residents experience lower satisfaction due to escalating housing costs and social isolation. A finer MGWR analysis uncovers spatial mechanisms that conventional regression models fail to detect.

As shown in Figure 4(A3), financial pressure suppresses life satisfaction more intensely in administrative boundary areas (coefficient: −0.32) than in city centers (−0.11) with an increase of approximately 191%. This spatial heterogeneity carries significant theoretical implications: (1) Administrative barriers disrupt public service distribution, amplifying financial strain through institutional costs such as social security mismatches for cross-district workers and inflated school district housing prices. (2) Mismatched residential–employment spatial patterns push commuting costs onto peripheral areas, creating “spatial buffer zones”, a trend particularly evident in the historic core of Gusu District, where traditional housing stock remains intact. Unlike prior studies that oversimplify economic factors as linear influences, this research employs spatial heterogeneity modeling to systematically analyze the geographically differentiated impacts of financial conditions on life satisfaction, laying the groundwork for subsequent subjective–objective analyses.

The built environment plays a critical role in shaping life satisfaction. Figure 5 reveals significant spatial heterogeneity in the impact of landscape quality, housing price, property service, business facilities, and housing type. Landscape quality exhibits a strong positive correlation within the inner ring, gradually weakening toward the periphery, suggesting that urban areas with well-maintained green spaces enhance satisfaction. Property service generally correlates positively, although some remote industrial areas show negative correlations, possibly reflecting disparities in service quality and resource allocation. Business facilities follow a similar trend, with positive correlations in urban cores but negative ones in peripheral zones, where commercial amenities lag behind. Housing type negatively correlates with satisfaction, particularly in suburban areas, highlighting the challenges of housing uniformity and limited options.

The commuting environment also significantly influences life satisfaction. As shown in Figure 6(C1), motorized commuting time exhibits a strong negative correlation with satisfaction. However, in cross-administrative areas, the coefficient polarizes—reaching −0.37—with a spatial heterogeneity intensity of q = 0.43. This phenomenon uncovers an often-overlooked “commuting institutional cost”: residents in border areas face implicit expenses such as fragmented social security and school district barriers (with education resource disparities up to 0.61), which amplify the negative effect of commuting time by 106%.

Although public facilities generally correlate positively with satisfaction (Figure 6(C2)), an anomalous negative association emerges in peripheral scenic zones. Geographical detector analysis reveals a pronounced “functional alienation” in these areas, where the spatial matching between the service radius of commercial facilities and resident demand is only 0.32. Consequently, tourist-driven facility allocation results in basic healthcare accessibility being 58% lower than in inner-city regions, thereby creating a “siphon effect” that compresses essential community resources.

In the Gusu area, an improvement in transport accessibility (+15%) paradoxically coincides with a decline in satisfaction (Figure 6(C3)). A spatial lag model attributes this to “commuting distance elasticity”—where transportation upgrades induce suburban sprawl, increasing the average commuting distance by 2.7 km and validating the “induced demand trap” in traffic dynamics. Additionally, environmental safety shows a robust positive effect in Gusu but only a weak correlation in migrant-concentrated zones (Figure 6(C4)). This discrepancy likely arises from a “visibility bias” in safety resource allocation—where a negative spatial association between surveillance density and population mobility results in a misalignment between actual safety improvements and perceived security among vulnerable groups.

It is noteworthy that we selected the MGWR model (

Table 6. Comparison of results between MGWR and GWR based on subjective data.

Model	R2	AICc
MGWR	0.530	3656
GWR	0.517	3688

) for its flexibility in adjusting bandwidth to accommodate local characteristics. Through iterative bandwidth selection, we secured a higher R² and a lower AICc. Specifically, MGWR outperformed GWR, achieving an R² of 0.530 compared to 0.517 and an AICc of 3656 versus 3688. These results confirm the efficacy and superiority of MGWR in capturing the spatial variability of life satisfaction, thereby providing more precise and reliable insights.

After conducting the MGWR analysis and visualizing the results, we compared the data with OLS outcomes (

Table 7. Comparison of OLS and MGWR analysis results of subjective questionnaire data.

Criterion	OLS Coefficients	MGWR Coefficients
	Mean	Mean	Min	Max	Bandwidth
Intercept	0.9896	−0.021	−0.321	0.420	150
Residence duration	0.1022	0.067	−0.462	0.454	97
Housing type	−0.0694	−0.042	−0.296	0.639	150
Housing price	0.0804	0.134	−0.267	0.363	150
Property service	0.0790	0.106	−0.167	0.445	149
Landscape quality	0.1090	0.061	−0.201	0.272	150
Business facilities accessibility	0.0945	0.061	−0.375	0.362	150
Financial condition	−0.0370	−0.076	−0.343	0.087	150
Public facilities	0.1099	0.205	−0.139	0.587	150
Environmental safety	0.1047	0.059	−0.202	0.380	123
Transportation mode	0.0796	0.103	−0.203	0.419	140
Motorized commuting time	−0.1255	−0.074	−0.860	0.426	86
R2	0.269	0.530 3656
AICc	3871

). The R² represents the proportion of variance in the dependent variable that can be explained by the model, while the AICc is used for model comparison, where lower values indicate a better fit to the data. The findings highlight that MGWR outperforms OLS. MGWR’s R² stands at 0.530, significantly higher than OLS’s 0.269, meaning MGWR explains a greater proportion of the variance in life satisfaction. Moreover, the AICc for MGWR (3656) is notably lower than OLS (3871), confirming that MGWR offers a superior fit for these data. These results underscore the importance of spatial heterogeneity and the need for more nuanced, location-sensitive modeling approaches when examining life satisfaction. By capturing spatial variations, MGWR provides a more precise understanding of how life satisfaction is shaped by different factors across regions, reinforcing its methodological advantage over traditional OLS models.

4.2. MGWR Results Based on Objective Geospatial and Social Media Data

4.2.1. Spatial Correlation Test and OLS Regression Results Based on Quantitative Data

Table 5 indicates a significant spatial agglomeration effect in the difference between 2020 and 2022 in the Weibo sentiment map. Meanwhile, Moran’s I index is more significant than zero, indicating a positive correlation. This means that communities with more substantial increases in the Weibo index tend to cluster together, while communities with lower increases in the Weibo index also tend to cluster together.

The results reveal significant spatial clustering in residents’ life satisfaction, as demonstrated by the positive spatial autocorrelation indicated by the Global Moran’s I index (

Table 8. Global Moran Index based on geospatial and social media data.

Dependent Variable	Moran’s Index	Z Value	p Value	E(I)
Life satisfaction	0.0934	10.7956	0.0000	−0.0006

Notes: Z value is the standard deviation difference between the regression coefficient and zero. p value is used to evaluate whether the regression coefficient is significantly different from zero. E(I) represents spatial autocorrelation.

). This clustering effect suggests that communities with higher life satisfaction tend to group together. The index’s significance level of −0.0006 (p < 0.05) confirms the robustness of this relationship.

The OLS regression retained seven key variables for further analysis using MGWR (

Table 9. OLS results with geospatial variables.

Variables	Coefficient a	StdError	t-Statistic	Probability b	Robust_SE	Robust_Pr b	VIF c
Intercept	3.6296	0.0709	51.1536	0.0000 ***	0.0743	0.0000 ***	N/A
Catering facilities	0.0005	0.0002	2.5461	0.0109 *	0.0001	0.0018 *	2.3249
Scenic spots	0.0084	0.0021	3.8905	0.0001 ***	0.0017	0.0000 ***	1.2818
Shopping facilities	0.0002	0.0001	2.1067	0.0352 *	0.0001	0.0070 **	2.0411
Medical service	−0.0031	0.0013	−2.2838	0.0225 *	0.0012	0.0111 *	2.1256
Housing price	0.0000	0.0000	3.3359	0.0008 **	0.0000	0.0002 **	1.1956
Green space ratio	−0.3410	0.1704	−2.0008	0.0455 *	0.1738	0.0499 *	1.0148
Population density	0.0000	0.0000	−1.5056	0.1323 *	0.0000	0.0693 *	154.7686
Weibo check-in	0.0006	0.0001	3.5432	0.0004 **	1.7321	0.0001 ***	1.1319
Entertainment check-in	−0.0009	0.0002	−3.4053	0.0006 **	0.0002	0.0003 **	154.6832

*** p < 0.01, ** p < 0.05, * p < 0.1. Coefficient ^a: the sign of the coefficient ^a indicates the direction of the influence (positive or negative). Probability ^b: this is the p-value corresponding to coefficient a. Robust_Pr ^b: this is the p-value of coefficient a calculated considering heteroskedasticity and autocorrelation. VIF ^c**: This indicator tests whether there is multicollinearity between independent variables.

). The findings show a positive correlation between life satisfaction and variables such as catering facilities, scenic spots, shopping facilities, and housing prices, reinforcing the idea that convenience and comfort directly enhance QoL. Interestingly, a higher green space ratio was negatively correlated with life satisfaction, a counterintuitive result that suggests potential underlying factors. However, due to the low R² of the OLS model, these relationships warrant deeper investigation, as OLS provides a limited fit, only reflecting the overall linear trends.

4.2.2. MGWR Results via Objective Geospatial and Social Media Data

The MGWR analysis, based on subjective survey data, reveals key spatial relationships between life satisfaction and urban factors. Six variables—catering facilities, scenic spots, shopping facilities, medical services, green space ratio, and housing prices—are retained for further examination. The results offer nuanced insights into spatial clustering and the varied impacts of these factors on residents’ well-being. Catering facilities (Figure 7(D1)) show a generally positive spatial correlation, with minor fluctuations. A transition from positive to negative correlation is evident when moving from the inner to middle urban rings, while a southern section of Xiangcheng district demonstrates a positive relationship. Housing prices (Figure 7(D2)) also display a stable yet varied spatial effect, with a positive overall correlation. However, certain areas, such as the junction between Gusu and Huqiu districts and parts of eastern Wuzhong, reveal negative correlations, possibly due to a lack of supporting infrastructure in high-cost areas, which diminishes life satisfaction.

The distribution of scenic spots (Figure 7(D3)) offers mixed spatial correlations. Northern Wujiang and southern Xiangcheng are positively associated with life satisfaction, whereas Gusu, densely packed with scenic locations, shows a negative correlation. This suggests that overcrowded tourist areas may detract from residents’ QoL, while balanced development yields more favorable outcomes. Medical services (Figure 7(D4)) present a mostly negative spatial correlation, potentially due to uneven resource distribution, which lowers satisfaction in underserved regions. Shopping facilities (Figure 7(D5)) exhibit both positive and negative spatial effects. Western areas near the industrial park border and sections of Suzhou Industrial Park have negative correlations, possibly reflecting incomplete commercial development in these transitioning zones. In contrast, central parts of the industrial park show positive associations, indicating that more developed areas contribute to higher satisfaction. Finally, the green space ratio (Figure 7(D6)) shows mixed results. Regions like eastern Wuzhong, northern Wujiang, and western Suzhou Industrial Park reveal negative correlations, likely due to poor accessibility and environmental quality compounded by high population density, leading to lower life satisfaction despite higher green space ratios.

These findings underscore the complex spatial dynamics in urban environments, where factors such as infrastructure and the spatial distribution of amenities can variably affect residents’ well-being. Further analysis using MGWR outperforms OLS by yielding a higher R² of 0.7626, offering a clearer explanation of the determinants influencing life satisfaction. This demonstrates that while overall patterns exist, the local context significantly shapes the impact of each variable.

In analyzing social media space metrics, three key factors—population density, Weibo check-ins, and entertainment check-ins—emerge as significant influencers. As shown in Figure 8, population density presents a mixed spatial correlation, with minor variations. The spatial pattern is divided by a Z-shaped zone stretching from eastern Huqiu to the northern inner ring and west of the industrial park, where a negative correlation is observed. This suggests that high density, coupled with overcrowding, inadequate infrastructure, or traffic congestion, fails to enhance life satisfaction. In contrast, the north and south of this divide show positive correlations, likely due to a more manageable population density in the middle ring, alleviating urban pressures. Weibo check-ins (Figure 8 (E2)) correlate negatively with life satisfaction. Frequent check-ins may consume significant time and energy, contributing to fatigue in real life and ultimately lowering overall well-being. This suggests that online engagement may come at the expense of offline QoL. In contrast, entertainment check-ins (Figure 8(E3)) demonstrate a positive spatial correlation with satisfaction, aligning with previous studies that show that leisure activities positively impact well-being.

The MGWR model demonstrates superior explanatory power in assessing life satisfaction. Its R² value reaches 0.495, outperforming both OLS and GWR (

Table 10. Comparison of results between MGWR and GWR based on quantitative data.

Model	R2	AICc
MGWR	0.495	4594.136
GWR	0.437	4689.265

), indicating a stronger ability to capture spatial heterogeneity in influencing factors. Additionally, MGWR achieves a lower AIC (4594.136) compared to GWR (4689.265), highlighting its advantage in balancing model fit and complexity. These findings underscore MGWR’s greater adaptability and precision in modeling life satisfaction dynamics.

The comparison between OLS and MGWR (

Table 11. Comparison of OLS and MGWR analysis results of objective spatial data.

Criterion	OLS Coefficients	MGWR Coefficients
	Mean	Mean	Min	Max	Bandwidth
Intercept	3.6296	0.175	−0.581	2.607	73
Catering facilities	0.0005	0.055	−0.950	0.725	70
Scenic spots	0.0084	0.066	−1.362	1.212	97
Shopping facilities	−0.0031	0.130	−0.258	0.744	130
Medical service	−0.3410	−0.034	−0.361	0.397	130
Housing price	0.0002	0.064	−0.781	0.482	130
Green space ratio	−0.0000	−0.024	−0.365	0.439	130
Population density	0.0000	−0.032	−0.848	0.647	126
Weibo check-in	0.0006	2.843	−2.587	22.253	60
Entertainment check-in	−0.0009	−2.410	−20.304	3.504	60
R2	0.049	0.495 4594.136
AICc	3346.347

) underscores the superiority of MGWR, which offers higher model fit and simplicity. These results highlight the importance of localized, spatially sensitive analysis for understanding urban well-being dynamics.

4.3. Nonlinear Influential Results via GBDT

4.3.1. Relative Importance of Life Satisfaction Factors

The relative importance of life satisfaction factors was assessed using a GBDT-based predictive model, achieving an explanatory power (R²) of 0.86. The factors fall into four broad categories: social demographics, built environment, social interaction, and commuting environment (

Table 12. The relative importance of predictors of life satisfaction.

Influence Factor	Prediction Weight	Level	Influence Factor	Prediction Weight	Level
Social Demographics (A)	10.83	/	Built Environment (C)	40.43	/
Housing price	7.91	2	Landscape quality	8.72	1
Household income	2.62	14	Property service	7.84	3
Household registration type	1.88	17	Business facilities convenience	5.82	6
Age	1.73	18	Educational facilities accessibility	5.59	8
Residence duration	1.68	20	Housing type	2.34	15
Educational background	1.65	21	Living units’ area	2.09	16
Family members	0.93	24	Dwelling height	0.12	36
Gender	0.34	31	Commuting environment (D)	43.63	/
Social Interaction (B)	5.11	/	Transport facilities	7.47	4
Psychological status	1.27	22	Accessibility Services	6.72	5
Financial condition	0.92	25	Environmental safety	5.59	7
Health condition	0.61	26	Mode of transportation	5.25	9
Exercise frequency	0.54	28	Traffic condition	5.11	10
Emotional stability	0.49	29	Transport greenery	3.20	11
Neighborhood relationship	0.42	30	Non-motorized commuting time	3.09	12
Job stability	0.28	33	Motorized commuting time	2.69	13
Working condition	0.22	34	Commuting distance	1.82	19
Emotional state	0.20	35	Commuting cost	1.19	23
Communication frequency	0.09	37	Commuting time	0.60	27
Time arrangement	0.04	39	Environmental aesthetic	0.31	32
Financial stability	0.03	40	Environmental cleanliness	0.07	38
R2	0.8636

). Notably, the commuting environment emerges as the most significant contributor, accounting for approximately 43.63% of life satisfaction variance. Within this category, transport facilities and accessibility services are the primary determinants. This underscores the pivotal role of urban mobility in shaping residents’ life satisfaction. The built environment follows closely, contributing 40.43% satisfaction, with residential conditions outweighing the impact of property services. Social demographics, with a combined importance of 10.83%, play a lesser role. Among them, household income stands out as the most influential factor, though its impact is limited compared to environmental variables. Lastly, social interaction contributes only 5.11%, indicating that its effect on life satisfaction is relatively minor in comparison.

4.3.2. Nonlinear Relationship Between Factors and Life Satisfaction

This session illustrates the nonlinear associations between social demographic variables and life satisfaction, focusing on six key factors such as household income and household registration type (Figure 9). Household income follows an inverted U-shaped curve, aligning with the Easterlin Paradox, which posits that while initial wealth increases happiness, further accumulation does not yield proportional gains [90]. This suggests that beyond a certain threshold, income fails to enhance well-being. Residence duration shows a fluctuating relationship with life satisfaction, peaking during 1–5 years, particularly in categories 3 and 4. Similarly, educational background exhibits an uneven impact, with no clear linear pattern emerging. Age, household registration type, and family size also display inconsistent trends, reinforcing the notion that these variables interact with complex socioeconomic factors. No single variable can fully explain the variations in life satisfaction, indicating the necessity of a more holistic approach to understanding these dynamics.

Figure 10 reveals a significant link between the built environment and life satisfaction, with factors such as landscape quality, housing type, and property services playing key roles. Both landscape quality and property services show a positive correlation with life satisfaction, though the rate of improvement slows in the middle range. This aligns with Maslow’s hierarchy of needs, where once basic requirements are met, satisfaction demands shift toward higher-order needs [91]. However, the relationship between living unit area and life satisfaction takes a different path. Initially, as living space increases, satisfaction rises. Beyond a certain threshold, the marginal gains in happiness diminish. Larger living spaces contribute to higher satisfaction, but this effect plateaus, suggesting that more space does not always equate to a better QoL.

Figure 11 illustrates the limited influence of social interaction on overall life satisfaction. Health condition and exercise frequency show slight fluctuations but remain consistently within the 3.5 to 3.8 range, supporting the “hedonic treadmill” theory. This theory posits that after adjusting to life events, individuals’ subjective well-being returns to a baseline [92,93]. Psychological status, financial condition, and emotional stability exhibit a marginally higher impact, but their effect remains between 3.5 and 3.75. This narrow range indicates that life satisfaction is influenced by broader dimensions, with psychological well-being emerging as a multidimensional construct not solely determined by social perceptions [94]. Moreover, factors such as neighborhood relationships, job stability, working conditions, and communication frequency have minimal effects. Notably, neighborhood relationships and communication frequency positively correlated with life satisfaction, aligning with social network theory [95], suggesting that the formation and maintenance of social ties significantly enhances subjective well-being.

In the scope of the commuting environment, transport facilities, accessibility services, and environmental safety emerge as the key drivers that influence life satisfaction (Figure 12). Both transport facilities and environmental safety show a positive correlation with life satisfaction. Efficient transport improves convenience, while a safer environment fosters a sense of security, collectively enhancing overall well-being. Better accessibility services also contribute positively to satisfaction. Transportation mode, traffic conditions, and greener transport options further boost life satisfaction, though the impact of greener transport shows diminishing returns. Initially, improvements in greenery lead to sharp gains in satisfaction, but beyond a certain threshold, additional greenery yields marginal benefits. This pattern aligns with the law of diminishing marginal utility, where initial enhancements deliver greater value than subsequent additions [96]. On the other hand, motorized commuting time, distance, cost, and overall commute time negatively correlate with life satisfaction. Long commutes drain energy and cause stress, ultimately reducing overall well-being [97].

5. Discussion

5.1. Subjective Perception and Objective Geospatial Data Result Comparison

The results reveal consistencies and significant spatial heterogeneity between subjective survey and objective geographic data in the MGWR model across four variable categories (

Table 13. Comparison of MGWR analysis results of the subjective and objective data.

	Subjective Questionnaire Data		Objective Geospatial Data
Items	Variable	Coeff_MGWR	Variable	Coeff_MGWR
Social demographics	Residence duration	0.067	Population density	−0.032
	Housing price	0.134	Housing price	0.064
Built environment	Property service	0.106	Catering facilities	0.055
	Housing type	−0.042	Scenic spots	0.066
	Public facilities	0.205	Medical service	−0.034
	Landscape quality	0.061	Green space ratio	−0.024
	Business facilities convenience	0.061	Shopping facilities	0.130
Social interaction	Financial condition	−0.076	Entertainment check-in	−2.410
			Weibo check-in	2.843
Commuting environment	Environmental safety	0.059	/	/
	Transportation mode	0.103	/	/
	Motorized commuting time	−0.074	/	/

). Regarding social demographics, housing prices demonstrate alignment between subjective perceptions and objective data, both positively influencing life satisfaction, underscoring their reliability as an indicator. Subjective residence duration also shows a positive impact, while objective population density exhibits a negative coefficient, indicating that higher population density likely diminishes satisfaction.

The built environment exhibits significant discrepancies in its impact on life satisfaction. While the convenience of business facilities aligns positively with both subjective and objective assessments, landscape quality presents a stark contrast. Residents’ subjective perceptions of landscapes are often optimistic, yet these perceptions are not always reflected in objective spatial data, suggesting that aesthetic qualities do not always translate into enhanced life satisfaction. Research has shown that although public green spaces provide psychological satisfaction, their actual contribution to life satisfaction is limited, whereas private green spaces, due to their higher service quality, yield more significant benefits [98]. Additionally, educational facilities accessibility and medical services show opposite effects: perceived convenience contributes positively to satisfaction, while actual medical infrastructure detracts from it. This mismatch highlights the regional variability captured by the MGWR model, reflecting the built environment’s diverse influence across different areas. Meanwhile, housing type shows a negative subjective impact, while property services exhibit a positive influence, implying that residents prioritize service quality over housing type when assessing landscape satisfaction.

For social interaction, the subjective perception of financial condition negatively affects satisfaction, while objective geographic data from social media, such as Weibo check-ins (0.988) and entertainment check-ins (−0.993), show divergent regional impacts. This variation suggests that while social interactions can enhance satisfaction in some areas, they may diminish it in others, pointing to the complexity of social dynamics across different urban settings.

In terms of commuting, residents perceive environmental safety and transportation options positively, associating safer environments and convenient transport with higher satisfaction. However, objective data reveal that longer motorized commuting times strongly correlate with reduced life satisfaction, reflecting the expected negative effects of extended commutes. Similar studies have highlighted that despite positive perceptions of commuting safety and convenience, excessive commuting significantly lowers life satisfaction [99]. Additionally, some studies suggest that commuting time alone does not significantly impact overall satisfaction [100]. For instance, research has shown that commuting mode, rather than time, may be a more significant factor in determining satisfaction levels [101]. This duality between perception and spatial reality underscores the nuanced role of commuting in urban life satisfaction. These findings emphasize the need to account for spatial heterogeneity when analyzing life satisfaction and suggest that subjective perceptions often diverge from objective spatial realities, particularly in complex urban environments.

5.2. Nonlinear and Prediction Factors Impact on Life Satisfaction

(1)

Nonlinear via GBDT

Nonlinear measurement results reveal critical insights for policy and practice. The analysis identifies the hierarchical significance of perceived factors impacting life satisfaction: built environment (40.43%) and commuting environment (43.63%) emerge as significant contributors, while social demographics (10.83%) and social interaction (5.11%) are comparatively less influential. These findings underscore the profound impact of objective environmental quality on subjective perceptions, indicating that individual experiences are strongly shaped by the surrounding environment. Furthermore, elements related to the built and commuting environments exhibit nonlinear and threshold effects, providing valuable reference points for urban planners.

Threshold effects are particularly notable in the built environment, where increased accessibility to facilities, convenience, landscape quality, and property services correlate positively with life satisfaction. However, a specific threshold exists for living space: optimal satisfaction arises from approximately 120 square meters. This nuance refines existing knowledge regarding housing conditions and their relationship with satisfaction [102]. Similarly, within social demographics, household income reveals a peak range (25,000–45,000 RMB), reinforcing the Easterlin Paradox in economics. This paradox suggests that while wealth accumulation can enhance happiness, further financial gains yield diminishing returns in life satisfaction [90]. Examining the commuting environment, nonlinear relationships reveal that most factors show significant regional variations. Even minor elements, such as aesthetic and accessibility satisfaction, exhibit threshold effects. This indicates that, similar to the built environment, objective commuting conditions significantly shape perceptions of commuting experiences and overall life satisfaction.

The nonlinear impact of factors such as transport greenery on life satisfaction exhibits an initial rapid increase in satisfaction followed by a slowdown. This pattern aligns with the law of diminishing marginal utility, suggesting that initial improvements in green spaces yield substantial satisfaction gains. However, beyond a certain level of greenery, additional enhancements produce progressively smaller benefits [96]. Therefore, urban managers should prioritize creating safe, comfortable, and convenient environments while adopting a balanced approach to green space management, adhering to the principle of moderation. Moreover, the gradual deceleration of life satisfaction with landscape quality aligns with Maslow’s hierarchy of needs. Once fundamental needs are met, individuals’ demands for satisfaction increase [91].

(2)

Prediction via XGBoost

Based on XGBoost prediction results (

Table 14. The performance of XGBoost predictions.

Evaluation Variables		R2	RMSE	MAPE
Subjective	Housing price	0.18	29.63	7.36
	Property service	0.64	17.70	3.72
	Business facilities convenience	0.20	31.50	6.43
	Financial condition	0.11	73.68	6.89
Objective	Housing price	0.27	11.50	3.11
	Catering facilities	0.36	11.85	2.98
	Shopping facilities	0.28	11.49	3.06
	Entertainment check-in	0.25	18.94	4.81

), our study further examines the congruence and divergence in the predictive power of subjective and objective factors on life satisfaction. In the subjective model, property service emerges as a standout predictor (R² = 0.64, RMSE = 17.7, MAPE = 3.72), underscoring the strong explanatory power of residents’ perceptions of property service quality. Conversely, among objective factors, catering facilities (R² = 0.36) and shopping facilities (R² = 0.28) show relatively robust predictive performance, although their overall R² values remain lower than those of subjective measures.

Notably, discrepancies exist between models for certain variables. For example, the subjective model yields an R² of 0.18 for housing prices, which is notably lower than the objective model’s R² of 0.27; moreover, the objective model reports a significantly lower error (RMSE = 11.5 vs. 29.63), suggesting that objective housing price data (e.g., transaction prices, supply–demand indices) offer more stable predictive insights than residents’ subjective evaluations.

Furthermore, within the subjective framework, financial condition (RMSE = 73.68) and business facilities convenience (RMSE = 31.5) incur high prediction errors, likely due to individual cognitive biases or survey design limitations. In the objective model, entertainment check-in facilities register the highest error (RMSE = 18.94), potentially attributable to insufficient spatial data granularity or nonlinear threshold effects.

Crucially, the results reveal a complementary relationship between subjective and objective models. For instance, while catering facilities in the objective model (R² = 0.36) and property service in the subjective model (R² = 0.64) both robustly predict life satisfaction, they capture distinct dimensions of the phenomenon. This underscores the necessity of integrating both subjective perceptions and objective spatial data to fully elucidate the drivers of life satisfaction.

Using the XGBoost model, we further identified the predictive trends of both subjective and objective factors on life satisfaction (Figure 13 and Figure 14). In Figure 13 and Figure 14, the abscissa represents the consistent factors identified through GBDT and MGWR, which reflect the alignment between subjective and objective variables. The ordinate represents the predicted change in life satisfaction resulting from the influence of these factors, quantified through perception evaluation.

Figure 13 displays a color-coded map generated via the SHAP (SHapley Additive exPlanations) algorithm, which quantifies the nonlinear influence weights of each factor. Larger color domains (e.g., housing price, financial condition) indicate stronger global influence, while the gradient from light to dark reflects the magnitude of impact—from low to high. For instance, although enhanced commercial facility convenience boosts satisfaction, excessively high housing prices fail to yield proportionate gains. Due to the clustering of subjective survey data, some variables (e.g., entertainment check-in) exhibit blurred boundaries, suggesting uneven data distribution or complex interactions; Figure 14 supplements these insights with a detailed scatter plot of predictions.

Figure 14 reveals that among subjective factors, property service satisfaction (MAPE = 3.72) and financial condition (MAPE = 6.89) show differing prediction accuracies, indicating that residents are more sensitive to service quality than to economic conditions. Conversely, among objective factors, housing price (MAPE = 3.11) and catering facilities (MAPE = 2.98) yield low prediction errors, affirming their stable predictive capacity. However, entertainment facilities (MAPE = 4.81) display a marked threshold effect, with their impact varying in stages as distance or density increases. The overall minor differences in prediction accuracy among objective variables likely stem from uniformly precise data.

5.3. Governance Mechanisms for Improving Life Satisfaction

Our findings underscore that subjective perception factors—particularly property service satisfaction—exert a much stronger influence on life satisfaction than objective spatial indices (R² = 0.64 vs. 0.36, with property service MAPE = 3.72). This indicates that residents prioritize “soft services” over mere hardware upgrades. We propose that property management should serve as a cornerstone of urban regeneration. Establish a dynamic evaluation system and implement periodic micro-updates (e.g., reducing green maintenance cycles to 15 days) to swiftly bridge the gap between residents’ expectations and their lived reality, thereby boosting life satisfaction [103].

For objective spatial factors, address facility distribution imbalances first. Our data reveal significant spatial thresholds for catering facilities (R² = 0.36) and medical resources (MAPE = 3.11). We recommend defining an 800 m service radius for community commerce and using property rights exchanges to remedy “facility lowlands”. Adopt a “15-min living circle—grid micro-center” model to ensure each grid unit hosts at least one integrated service center that offers basic medical and convenience services. In older communities, implement innovative “spatial functional overlap” strategies, such as converting idle power distribution rooms into shared wellness stations, to enhance space utilization and improve convenience.

Cognitive gaps between subjective and objective factors demand effective feedback mechanisms. Develop a “community demand heatmap” platform to monitor real-time complaint hotspots, thereby fostering positive interactions among the government, property managers, and residents. Host “update workshops” that engage residents in contentious issues (e.g., parking space redesign) to boost policy acceptance and satisfaction. Iterative feedback mechanisms have been shown to increase facility renovation satisfaction by 23% compared to traditional public announcement systems. Additionally, establish a “policy effect tracing system” to dynamically adjust interventions, ensuring long-term policy efficacy.

Beyond material upgrades, our study emphasizes building community belonging and cultivating cultural capital [64]. We propose creating a “community time bank” to encourage skill exchanges that strengthen social cohesion. Furthermore, a “third space activation fund” should be established to support semi-public spaces—such as community cafés and bookshops—thereby enhancing informal social networks and community social capital [104].

Research on nonlinearity and nonstationarity further highlights the importance of threshold effects and spatial heterogeneity, underscoring that urban regeneration is a long-term, dynamic process [65]. To ensure the sustainability of regeneration outcomes, we recommend establishing a “whole-life cycle management framework”. This framework should include a dedicated community renewal trust fund to reinvest land value gains into regeneration and operations and a facility usage efficiency monitoring system that triggers automatic adjustments when usage falls below a defined threshold. Such a dynamic operational model will extend the vitality of public spaces and secure the long-term stability of urban renewal efforts.

6. Conclusions

This study investigates the impact mechanisms through which residents’ subjective perceptions and built environment affect life satisfaction, focusing on Suzhou. It presents several innovative contributions to the field. First, in terms of perspectives, this research goes beyond conventional single-dimensional analyses by integrating both subjective and objective factors to uncover the intricate interactions and mechanisms affecting life satisfaction. This multi-dimensional approach provides a more comprehensive theoretical framework. Additionally, the use of MGWR enables a deep exploration of spatial heterogeneity, revealing significant variations in life satisfaction across different urban areas, thereby offering precise spatial planning recommendations for urban renewal.

Methodologically, this study innovatively combines GBDT, MGWR, and XGBoost to analyze the complex relationships between subjective and objective factors. The integration of GBDT for nonlinear threshold analysis and XGBoost for model optimization significantly enhances the predictive accuracy of the model. This multi-methodological approach addresses the limitations of conventional methods by capturing the nonlinear and spatially heterogeneous nature of the relationships between subjective perceptions, the built environment, and life satisfaction, offering a novel perspective for future research.

Through comprehensive analysis, the study identifies key factors influencing life satisfaction, including landscape quality, housing prices, property services, transportation modes, environmental safety, catering facilities, and scenic spots. Notably, the research reveals inconsistencies in the influence of certain factors, such as green space quality and green space ratios, highlighting the need for a human-centric approach in urban development that prioritizes both quantity and quality. Furthermore, the study demonstrates the nonlinear relationship between subjective perceptions and the objective environment, showing that factors like greenery levels do not consistently enhance life satisfaction but vary based on specific conditions. This counters the oversimplification of linear relationships in previous studies, providing a more nuanced understanding of life satisfaction dynamics.

By focusing on spatial heterogeneity within urban contexts, this research addresses a significant gap in prior studies that often overlooked the spatial dimension of life satisfaction. The findings underscore how variations in life satisfaction across different geographical areas and community environments are influenced by specific environmental conditions, offering valuable insights for policymakers and urban planners.

Nevertheless, our study acknowledges limitations, notably in sample size and scope. Future research should expand the sample, incorporate longitudinal data, and compare multiple cities to further enhance the generalizability of these findings. Furthermore, the reliance on Weibo check-in data may not fully represent local resident behavior, as non-local users are also included. Future research should complement Weibo data with additional local sources to enhance representativeness and robustness. Such efforts will contribute to the development of high-quality urban environments that effectively balance subjective well-being with objective environmental quality.