SmartPrice: A Multi-modal, Explainable AI System for Predicting House Prices Using Structured Data and Property Descriptions

Accurate residential property valuation is important to real-estate investment mortage approval, taxes and urban planning. Conventional house price prediction systems rely primarily on structured numerical attributes and human expertise, often ignoring unstructured property descriptions that contain valuable semantic indicators such as renovation quality, interior features, and neighborhood appeal. This paper proposes SmartPrice, a multi-modal explainable artificial intelligence framework that integrates structured tabular features with textual property descriptions for enhanced housing price prediction. The proposed architecture employs a dual-branch learning pipeline: structured numerical and categorical attributes are modeled using XGBoost, while unstructured text descriptions are encoded through a one-dimensional Convolutional Neural Network. The extracted representations from both modalities are fused at the feature level and supplied to a regression layer for final price estimation. To ensure transparency and trustworthiness, SHAP is incorporated to provide both global and local interoperability of model predictions. Experiments conducted on the Dataset collected from Kaggle [1] show that our proposed Multi-model framework surpasses traditional regression models and stand-alone machine learning approaches in terms of predictive performance. SmartPrice obtains an R2 score of 0.92 and reduces Root Mean Squared Error by approximately 10–14% with respect to tabular baselines. The system is launched as a real-time web application using Streamlit [2], enabling interactive predictions and feature-level explanations for end-users. The results confirm the effectiveness of Multi-modal learning combined with explainable AI for real-estate analytics and highlight its potential for practical deployment in automated valuation systems.