1# RECORD 871e09a43b07cddac0d3c99559043f1bd6c1a6142# TITLE: Feature Attribution and Error Diagnosis of a Random Forest Housing Price Prediction Model Integrating SHAP and LIME3# YEAR: 2026.0 VENUE: Theoretical and Natural Science DOI: 10.54254/2753-8818/2026.dl344094# AUTHORS: Ruoxiang Wang5# PAGES: 8 CANDIDATE TABLE PAGES (1-based): []6# ABSTRACT: Housing is one of the most basic needs of humanity globally, and fluctuations in housing prices have garnered significant attention. Currently, the scale of China's second-hand housing transaction market is rapidly expanding. Understanding the factors influencing second-hand housing prices can enhance transaction efficiency and achieve a balance between supply and demand. This article obtains the Ames dataset from Kaggle and selects the Random Forest (RF) model, which is a relatively mature model in the field of housing price prediction, for training. It integrates Shapley Additive exPlanations (SHAP) and the Local Interpretable Model-agnostic Explanation (LIME) to analyze the importance of housing features. The results indicate that the RF model exhibits good performance in predicting housing prices. The SHAP analysis reveals that the most significant global features influencing second-hand housing prices are the quality of building materials and decoration, followed by living area, and there is an interaction between these two factors. Furthermore, it utilizes LIME to perform local attribution on high-quality small-area samples in the test set. The results revealed that the local contribution of living area in such samples was significantly reduced, and replaced by features such as housing functionality, indicating that the decision logic of the RF model may deviate in local areas that deviate from the mainstream distribution.789===== PAGE 1 (score 8, numeric-paren 0, table-words 0, var-words 8) =====1011Proceedings of CONF-MPCS 2026 Symposium: Theoretic Physics and Plasma Physics12DOI: 10.54254/2753-8818/2026.DL3440913© 2026 The Authors. This is an open access article distributed under the terms of the Creative Commons Attribution License 4.014(https://creativecommons.org/licenses/by/4.0/).153116Feature Attribution and Error Diagnosis of a Random Forest17Housing Price Prediction Model Integrating SHAP and18LIME19Ruoxian Wang20Institute of Science, Minzu University of China, Beijing, China21chordwangruoxian@outlook.com22Abstract. Housing is one of the most basic needs of humanity globally, and fluctuations in23housing prices have garnered significant attention. Currently, the scale of China's second-24hand housing transaction market is rapidly expanding. Understanding the factors influencing25second-hand housing prices can enhance transaction efficiency and achieve a balance26between supply and demand. This article obtains the Ames dataset from Kaggle and selects27the Random Forest (RF) model, which is a relatively mature model in the field of housing28price prediction, for training. It integrates Shapley Additive exPlanations (SHAP) and the29Local Interpretable Model-agnostic Explanation (LIME) to analyze the importance of30housing features. The results indicate that the RF model exhibits good performance in31predicting housing prices. The SHAP analysis reveals that the most significant global32features influencing second-hand housing prices are the quality of building materials and33decoration, followed by living area, and there is an interaction between these two factors.34Furthermore, it utilizes LIME to perform local attribution on high-quality small-area35samples in the test set. The results revealed that the local contribution of living area in such36samples was significantly reduced, and replaced by features such as housing functionality,37indicating that the decision logic of the RF model may deviate in local areas that deviate38from the mainstream distribution.39Keywords: second-hand housing prices, random forest, SHAP method, LIME, price forecast401. Introduction41In the research on the influencing factors of second-hand housing prices, traditional methods mostly42involve constructing multiple linear regression models for analysis. However, the real estate market43is a complex nonlinear system, and traditional linear models are sensitive to outliers in data. House44price data with skewed distributions can affect the robustness of the model.45Currently, the new generation of artificial intelligence, as a digital tool, is thriving and has46become an effective way to enhance the accuracy of real estate appraisal. Based on machine learning47and modeling, Hong complements traditional regression methods by studying the linear or nonlinear48relationship between dependent and independent variables, as well as the hierarchical structure of49factors determining housing prices [1]. Shi conducted training and prediction of second-hand50