Využití metod strojového učení pro predikci cen bytů

The aim of this thesis is to apply various machine learning methods to predict apartment prices (on-line valuation) and to compare their predictive performance in the Czech housing market using standard evaluation metrics such as RMSE and MAPE. The dataset used for training and testing the models contains market property price estimates prepared by professional appraisers in the time of mortgage application during the past year. These data were provided in unified form by the largest mortgage lenders operating in the Czech Republic through Lux Property Index, Ltd. In addition to comparative analysis, the thesis also offers a more detailed examination of the importance of apartment characteristics as predictors of apartment prices using multiple regression analysis. The results will not only make it possible to assess which of the selected models demonstrates the highest predictive performance for apartment valuation but will also contribute to a deeper understanding of the factors that mostly influence apartment prices.