Estimation of the Probability of Default on Bank Loans

The purpose of this study is to find the best way to predict credit risk, expressed as the probability of default. It compares different mathematical models such as logistic regression, decision trees, gradient boosting over decision trees and neural networks. The main source of data will be “Spark-interfax” which has a lot of information about a particular firm, such as financial statements, internal credit risk assessment, identification of affiliates, etc. The data on the credit history of the company will also be used. By the end of the study, the best model within the above-mentioned models will have been selected in terms of forecasting quality. In addition, the strongest variables contributing to the model's accurate forecasting will be identified, which will help understand the picture of the company's financial solvency as a whole.