Application of Machine Learning for Real Estate Market Analysis and Prediction of Property Prices

The main question of the research was: Which machine learning model most accurately predicts the cost of residential real estate objects? The goal of the research was to conduct an in-depth analysis of the real estate market, including the quantitative influence of various factors in monetary terms, estimated using machine learning algorithms such as decision tree, random forest, extreme gradient boosting, fully connected layer and stacking ensembles. Changes in real estate prices have always been of interest to economic agents. The Central Bank needs to be informed about movements in this market, as delayed reactions can lead to inflation. Homeowners wanting to sell their property fear underpricing, while buyers fear overpaying. Despite the growing trend of applying machine learning algorithms to real estate market analysis, there are not many works conducted within the territorial framework of Russia. In addition to a broader methodology, the validity of the results was enhanced by using a larger number of accuracy metrics (MAE, MAPE, and RMSLE), as well as time series cross-validation for optimizing models and selecting the best hyperparameters, with the aim of generalizing the results for their further application. As a result, the best approach to predicting prices on the St. Petersburg real estate market is the stacking ensemble of extreme gradient boosting and a fully connected neural network. The obtained conclusions have important implications for stakeholders in the St. Petersburg real estate market, including buyers, sellers, investors, and macroeconomic agents such as the Central Bank and the Government of the Russian Federation. By providing an approach to property valuation based on data, the study can contribute to making more informed decisions in this sphere.