Predikce cen nemovitostí v ČR

This thesis focuses on the prediction of real estate prices in the Czech Republic, which is a key topic due to the high proportion of property owners and frequent cases of overpaying when buying real estate. The main objective is to apply and evaluate different methods of real estate price prediction in the Czech housing market, using machine learning techniques. The thesis gradually covers the history and development of prediction methods, their theoretical foundations to practical applications, with emphasis on proper data collection and cleaning for prediction purposes. The analysis of different predictive models reveals that XGBoost-based methods, especially when combined with advanced techniques for feature selection and hyperparameter optimization, offer the best results. The XGBoost regressor model combined with RandomizedSearchCV and the SHAP method achieved the lowest mean absolute error (459770 CZK). This result was supported by visual analysis, which showed high accuracy of the model in predicting property prices up to CZK 4 million, with some deviations for higher price categories. TensorFlow was also found to provide comparable results to some traditional models. The work contributes to a better understanding and prediction of property prices in the country, which is crucial for buyers, sellers and investors in the real estate market.