Decoding Rent Determinants in Urban Housing Markets: A Multi-Perspective Multimodal Machine Learning Analysis

Urban housing rents are central to socioeconomic dynamics and urban sustainability, shaping affordability and quality of life. Existing research largely relies on linear models and focuses on economic, demographic, and locational factors, often neglecting complex nonlinear interactions and the impact of human perceptions. This study introduces a comprehensive, multi-perspective framework that integrates housing attributes, living convenience, competition, location, accessibility, and quantified perceptual metrics using multimodal machine learning. Advanced techniques, including XGBoost, SHAP, Partial Dependence Plots (PDPs), Interpretative Structural Modeling (ISM), and Bayesian Network (BN), capture nonlinearities, interactions, and hierarchical dependencies among rent determinants. Housing attributes and living convenience indicators exert the strongest cumulative influence on rents, while perceptual variables rank third, providing significant, threshold-dependent contributions and explaining up to 21.66% of rent variation. Notable interactions are identified between accessibility, facility density, and perceptual quality. The ISM–BN analysis uncovers multi-level pathways, demonstrating how both environmental features and human perceptions jointly influence rents. This framework offers actionable insights for equitable housing and urban planning policies, supporting data-driven decisions in complex urban rental markets.