A Spatial Regression Approach in Property Valuation Using Machine Learning

This work aims to model the sales of housing prices in Acapulco City using a spatial approach. For this purpose, we considered different models that change their explanatory variables; we used variable selection methods, such as topological data analysis and correlation analysis, to compare and choose a better model. Machine learning algorithms were trained with distinct models to learn and extract the relationships between the explanatory variables and the hedonic price. Using non-parametric statistics, we selected the best algorithm and model in terms of its coefficient of determination and computational cost. A random forest algorithm was selected, with which a spatial approach, namely a Kriging regression, was addressed to predict housing prices. Our final model can predict such prices in locations non-sampled as spatial interpolation.