Improving Housing Price Prediction with Machine Learning: Insights from Yogyakarta

Abstract

Predicting housing prices accurately remains a major challenge in the real estate industry, especially in fast-urbanizing areas where both structural and locational factors are at play. This study focuses on the Special Region of Yogyakarta, Indonesia—a city with varied land use and dynamic housing market conditions—to explore how machine learning (ML) can support better price forecasting. Using the CRISP-DM framework, we analyzed data from 2,020 residential listings, incorporating variables such as building area, land size, number of rooms, and district. Among the several classification models tested, Random Forest achieved the highest accuracy and F1-score. According to the feature importance analysis, building area, land area, and district emerged as the strongest predictors, while vertical features like the number of floors had relatively little effect. These findings suggest that in Yogyakarta’s market, observable physical features may play a bigger role in price determination than location-specific factors. While the study offers a practical framework that real estate professionals can apply, its reliance on structural data and a single-region focus does limit how broadly the findings can be applied. Future research could expand on this by including socioeconomic or environmental variables to strengthen model performance and relevance across different markets.