How it works 

Our AVM is trained to determine current market values ​​of one or more real estate objects. Machine Learning (ML) technology is used for this, which is fed with various information about the object to be valued (such as year of construction, living area, plot area, energy label and housing type), location characteristics, neighborhood data, market developments and comparable transactions. The data we use for this go back to 1995.

Quality assurance

When calculating a market value our AVM assumes normal sales conditions. For example, only private transactions are considered. Details such as family transactions, foreclosure sales or other outliers that could cast doubt on the representativeness of a house value are not included in our model.

To guarantee the very high quality of our AVM it is necessary that the model is continuously provided with current and reliable data. We achieve this by ‘retraining’ our model monthly with the latest real estate data. To do so, we randomly split our dataset into a training and validation set. The validation set is used after the training to avoid any sample bias. We compare the predictions to the actual transaction prices and calculate various statistics such as the mean, the median deviation and the spread coefficient.