Forecasting pandemic-induced changes in real estate market values through machine learning approaches

In this study, a new temporal segmentation method is used to forecasting the real estate market based on the structural and spatial attributes of 676 houses in Niğde, Türkiye, from the years 2019 to 2022. Artificial Neural Networks (ANN), Random Forest (RF), Decision Tree (DT), and K-Nearest Neighbours (KNN) were employed for model development and comparative performance analysis. According to the results, the ANN model that used temporal variables showed the most successful performance by achieving the highest R2 for 2019 (1. period: 0.979, 2. period: 0.990, 3. period: 0.914, 4. period: 0.831) and 2022 (1. period: 0.971, 2. period: 0.975, 3. period: 0.586, 4. period: 0.896) scores. Additionally, the COD values (5%–10%) and PRD values (0.98 to 1.03) remained within the acceptable range, further validating the model’s reliability. RF model showed more effective performance than other models by achieving the highest R2: 0.510 for 2019 and R2: 0.509 for 2022 when temporal variables were excluded. These findings highlight the importance of integrating time-sensitive parameters into valuation models to improve forecast accuracy and robustness. The study offers a replicable, flexible methodology for crisis-responsive valuation, providing valuable insights for policymakers, investors, and urban planners aiming to mitigate risks and enhance resilience in real estate market decision-making.