Objective This study comprehensively explores the nonlinear influence mechanism of built environmental and demographic factors on metro and non-motorized transport willingness to interchange from a human-centric perspective, aiming to provide detailed design recommendations for optimizing the non-motorized transport environment and fostering the sustainable development of rail transit within transit-oriented development (TOD) projects. As urbanization accelerates at an unprecedented rate, the surge in vehicle ownership has exacerbated traffic congestion and air pollution, making the integration of "rail-bus-slow travel" networks a strategic imperative for sustainable urban mobility. However, rail transit systems often struggle to deliver seamless travel experiences due to their station layout characteristics, particularly in addressing the "last-mile" connectivity challenge, which hinders the overall efficiency and attractiveness of public transport. By constructing a micro-circulation connection system that combines rail transit with non-motorized transport, this research seeks to enhance the service coverage of rail stations, promote multimodal transport integration, and optimize the overall efficiency of urban transportation systems. Notably, existing studies predominantly focus on the interplay between socio-economic attributes and built environment factors on individual route and destination choices, yet there remains a significant gap in understanding the spatial interventions of non-motorized transport behaviors from an environmental perception , particularly in the context of rail transit connections. This study addresses this gap by adopting a human-oriented approach to unravel the complex interactions between demographic factors, built environment features, and travel behavior.

Methods Conducted in Chengdu, a pioneering city in TOD development with 423 metro stations across 16 lines, this study employs a multi-methodological approach to ensure the robustness and reliability of findings. A questionnaire, grounded in the Stated Preference method, was meticulously designed to capture both demographic factors and built environment features, encompassing variables such as gender, age, income, education level, environmental awareness, destination distance, bicycle accessibility, road speed limit, road continuity, and land use composition. To ensure data quality and representativeness, the D-optimal design methodology was utilized to generate 20 factorial combinations for scenario-based questioning, effectively capturing the complexity of real-world travel decisions. Data collection was facilitated through the Credamo online survey platform, with stringent filters applied to respondents’ geographic location, daily travel patterns, and historical questionnaire response rates, yielding 863 valid responses that reflect the diversity of Chengdu’s urban population. Data analysis was rigorous and multifaceted, employing Excel 2022 and SPSS 25.0 for descriptive statistical analysis to provide an overview of the dataset. Least absolute shrinkage and selection operator regression, a powerful machine learning technique, was leveraged to extract the weights of perception factors, enabling the construction of weighted perception indices that account for the relative importance of different environmental attributes. The study further explored the nonlinear characteristics of built environment variables by creating interaction terms between socio-economic and built environment variables, thereby capturing the complex interplay between individual characteristics and the built environment. Based on the parametric scale transformation principle, a linear equivalence method was employed to convert the 7-point scale to a 5-point scale for neural network model training, thereby enhancing overall model training accuracy.

Results The findings reveal a pronounced preference for walking as the primary mode of continuous transport, followed by bicycling, with motorized transport exhibiting the lowest willingness to choose, highlighting the potential for promoting active transportation modes in urban areas. Notably, female respondents demonstrated a stronger inclination towards non-motorized transportation modes, suggesting the importance of gender-sensitive design in urban planning. Income and walking distance have emerged as the primary determinants influencing the willingness to choose non-motorized modes of transportation. Analysis of interaction dependence plots in the SHAP (Shapley additive explanations) analysis reveals that individuals with higher incomes are more inclined to opt for non-motorized travel under scenarios involving longer distances or higher speed limits. Furthermore, varying land use mixes can either enhance the propensity to choose walking or reduce the likelihood of selecting motorized transportation. Complex nonlinear relationships were observed between walking distances and different demographic groups, with varying sensitivities to built environment factors across socio-demographic factors, underscoring the need for context-specific interventions. Furthermore, a notable exclusivity was identified between non-motorized and motorized transportation modes, as well as between walking and bicycling, particularly pronounced among female respondents, highlighting the importance of integrated transport planning that considers mode competition and complementarity.

Conclusion This research contributes to the literature by elucidating the nonlinear influence mechanism of built environment and socio-demographic factors on non-motorized transportation mode preferences from a human-oriented lens, thereby advancing theoretical frameworks for understanding travel behavior. Practically, the study proposes human-centric hierarchical optimization strategies, offering a scientific foundation for the refined design of non-motorized transportation systems in TOD projects, which can enhance the overall efficiency and attractiveness of public transport. These insights are instrumental in promoting the seamless integration of "rail-slow" networks, a critical step towards achieving urban transportation carbon neutrality and building sustainable, livable cities. However, the study acknowledges limitations, particularly regarding model explanatory power constrained by variable selection, suggesting future research could incorporate additional variables, such as individual attitudes towards sustainability and technological acceptance, to enhance predictive accuracy. Furthermore, the underrepresentation of low-income and low-education groups in the sample highlights the need for future studies to adopt a more dynamic approach, encompassing diverse social demographics across varying temporal and spatial contexts, to ensure the equity and inclusivity of transport planning.