Housing Price Prediction Using AI Agent With Causal Inference for Societal Factor Selection

Ray-I Chang, Dong-Ming Hsieh, Tsung-Sheng Lin

Abstract: Current housing price prediction (HPP) models face difficulties in identifying and explaining the causal impact of multidimensional societal factors. Existing statistical and deep-learning approaches rely heavily on correlation-based modeling, which limits their interpretability and adaptability to dynamic urban conditions. To address this gap, this study proposes an autonomous artificial intelligence (AI) agent framework that integrates a cyber-physical-social system (CPSS) ontology with a structural causal model (SCM) for intelligent societal factor selection and interpretable housing price prediction. AI agents operate through a self-adaptive cycle of perception, causal reasoning, and iterative learning. The CPSS ontology organizes complex societal variables into six core dimensions, enabling structured causal analysis, whereas SCM performs Bayesian-network-based factor screening. The selected factors are then used in an XGBoost model to enhance prediction accuracy and interpretability. The proposed CPSS-SCM framework provides quantitative evidence of the varying impacts of societal factors across urban environments, supporting transparent and region-specific real estate market analyses. Experiments conducted on 23,901 real estate transaction records from four districts in Taipei demonstrated significant improvements over traditional housing-factor-only models: the mean absolute percentage error (MAPE) decreased by 5%, the root mean squared error (RMSE) was reduced by 11%, and the coefficient of determination (R2) score improved by 4%. The R2 score consistently remained between 0.92 and 0.93 across districts, indicating excellent robustness. This integration represents a novel causal–predictive pipeline that improves both model transparency and predictive robustness in urban analytics.

External IDs:dblp:journals/access/ChangHL25