Dynamic imbalanced business credit evaluation based on Learn++ with sliding time window and weight sampling and FCM with multiple kernels

Abstract

A good model of business credit evaluation is an important tool for risk management. Although the dynamic imbalanced data flow is more consistent with the form of collected financial data in the actual situation, existing studies seldom research financial data as this form. This paper proposes a new ensemble model for dynamic imbalanced business credit evaluation based on the improved Learn++ and fuzzy c-means (FCM). To handle dynamic imbalanced financial data, Learn++ is improved by using a sliding time window (STW) and weight sampling (WS). This method is termed Learn++.STW-WS. STW can divide data with the same concept into the same dataset to solve the problem of concept drift which characteristic in dynamic data. Additionally, WS can redistribute the weights for samples of different classes to resolve the issue of imbalance. To satisfy the demand of Learn++.STW-WS on the prediction accuracy of a base classifier, FCM is improved by multiple kernels (MK), and is designated as MK-FCM. Several kernel functions are integrated to construct MK by the mean method, and MK is adopted to improve the calculation method of distances among points for FCM. Therefore, this new ensemble model can solve the problems of dynamic data and imbalanced classes at the same time. In the empirical research, financial data from Chinese listed companies are selected to evaluate business credit risk, and the associated models are adopted to make comparative analysis. The experiment results can fully demonstrate the good performance of the new ensemble model in terms of handling dynamic imbalanced financial data.