A local classifier with built-in tools for class imbalance correction and local model bagging

Borrower default prediction is one of the first problems [1] to be addressed using machine learning. This article describes a new integrated borrower default prediction method (hereinafter referred to as the integrated method), which is a local classifier with built-in tools for class imbalance correction and local model bagging. The development of the integrated method resulted in methods for sequential three-stage optimization of the integral method components and adaptive class imbalance correction.

Objective – to describe the integrated method and auxiliary methods for its configuration: an adaptive class imbalance correction method and a sequential optimization method. To demonstrate the effectiveness of the proposed methods.

On average, on five independent datasets, the proposed integrated method demonstrated an increase in F1-score of 0.395 +/– 0.051 and recall of 0.412 +/– 0.049 relative to the baseline model.

The practical applicability of this knowledge lies in reducing direct losses from loan defaults while improving the classification quality of borrower default prediction models and incorporating the results into the training programs of students in specialized programs.