Author page: Akhrameev Maxim D.

Forecasting bankruptcies of counterparties based on payment discipline data

DOI: 10.33917/mic-4.93.2020.47-56

In this article, we study the problem of forecasting bankruptcy of firms using data on payment discipline. Most previous researchers used the balance sheet as a data source, while data on payment discipline will reduce the time before making a decision on the firm, as well as obtain reliability ratings based on other types of data. To predict bankruptcy of the firms proposed a new method of work with highly unbalanced data, which consists in training the classifiers on the automatically generated sub-sample and averaging the obtained results. Random forest served as a classifier for subsamples, and AUC-score was used to check the quality of the model, which showed good results.