AUTOMATIC VALUATION OF RESIDENTIAL PROPERTIES USING MACHINE LEARNING MODELS

Recent literature on real estate valuation has provided evidence on the good performance of machine learning models in predicting price behavior, especially when compared to those obtained by traditional valuation methods. The latter are widely used in professional practice. However, some limitations are still highlighted, such as the black box design and the difficulty in interpreting the results provided by these techniques. This work aims to compare the results and performance of different machine learning models applied in the field of residential real estate valuation. For this purpose, a large database of property listings in the city of Madrid has been compiled, which allows the sample to be divided into training and test groups. The comparison between the models has been carried out through different metrics, among which the MAPE (Mean Absolute Percentage Error) stands out as one of the favorites of valuation companies. The metrics we have used confirm a good generalized performance for the set of trained models, with relatively small variations after the validation process.