Real Estate trading (renting, selling) is carried out every day, so the forecasting the value of Real Estate is very important. The aim of the master's thesis was to develop a forecast model for Real Estate valuation with data mining, which forecast the value of Real Estate (contract rent or price for apartment) based on data from various sources. An important factor in the preparation of data sets was the integration of data that indirectly affects the value of Real Estate. We extended the baseline REMR data set with additional data and created two new data sets - renting and buying apartments. We carried out cleaning procedures on these data sets (removal of outliers, imputation of missing values). We also carried out a feature selection. Using forecast methods (linear regression, random forests), we made data from the data sets forecast models for forecasting the value of Real Estate and evaluated them.
When forecasting contract prices for apartments, random forest defects reached the lowest mean absolute error (MAE) of € 10,986.15, which is better than with linear regression, where the MAE is € 14,496.75. Both methods exceed the MAE of € 25,424.58 of the null model. Also in forecasting contractual rents for apartments, random forests have obtained better results (MAE is € 61.57) than with linear regression (MAE is € 81.20), which is better than the null model (MAE is € 95.15).
The forecast model includes the state of the market and represents an alternative to the current MGRT evaluation, based on complex evaluation models.