A Machine Learning Based Enhanced Property Valuation Framework for Volatile Markets

Zimbabwe’s real estate sector is plagued by valuation inaccuracies due to hyperinflation, informal transactions and fragmented data infrastructure, making traditional valuation methods unreliable. This study was conducted to develop a machine learning-based valuation framework tailored to volatile and data-scarce property markets. The research addresses the core problem of valuation inconsistency by leveraging supervised learning models to improve predictive accuracy in such challenging contexts. Following the CRISP-DM methodology, a dataset of 1,500 property transactions from 2019 to 2024 was compiled, incorporating structural, geospatial and macroeconomic variables. Four models namely Linear Regression, Decision Tree, Random Forest and XGBoost, were trained and evaluated using an 80/20 data split and 10-fold cross-validation. Random Forest emerged as the most effective, achieving the highest predictive accuracy (MAE = 7,420.10; RMSE = 9,864.27; R² = 1.000), followed by XGBoost (R² = 0.88). These findings underscore the robustness of ensemble models in capturing nonlinear dynamics within unstable markets. The proposed framework presents a scalable, interpretable tool for automated valuation models (AVMs), offering practical implications for mortgage risk analysis, pricing strategies and policy formulation in Zimbabwe and similar emerging market contexts.