Personal credit risk identification based on combined machine learning model

Abstract

Personal credit risk assessment is essentially based on the classification of indicators or characteristic variables related to personal credit. Currently, a single model is often used to classify normal customers and defaulting customers. However, the single model may face the risk of a higher type II error rate, that is, defaulting customers are misjudged as normal customers. From the perspective of the combined model, this paper uses the confidence-weighted voting strategy to give a single model different weights to construct a C5.0-SVM combined model. The study finds that the total classification accuracy of the C5.0-SVM combined model is higher than that of the single model, and the type II error rate is lower than that of the single model.