Application of methods based on ensembles and deep neural networks to estimating the cost of commercial real estate

Abstract

The paper considers the possibility of using ensemble machine learning models and artificial neural networks to solve the problem of assessing the value of commercial real estate. There are some models such as the gradient boosting model and the TabNet model have been trained. The main goal of these models is predict the value of commercial real estate without creating dependencies between data by the analyst. The proposed solutions are considered from the point of view of the banking sector. The best predictive model is the gradient boosting model implemented using the LightGBM library. The advantages of this model are associated with its ability to "resist" the presence of outliers in the data and a low propensity for retraining.