Multi-classification assessment of personal credit risk based on stacking integration

Abstract

In the multi-category assessment of personal credit risk, the prediction accuracy of a single classifier is often low, and each category cannot be well distinguished. In this paper, the stacking ensemble algorithm is used to integrate the base classifiers to build a multi-class evaluation model. Through the empirical analysis of the five-class data of an anonymous commercial bank in China, six types of base classifiers are selected to train the data, and the evaluation results of each single classifier is obtained. The four classifiers with the best classification effect are used as the first-layer base classifier in the stacking ensemble, and Random Forest is selected as the second-layer meta-learner. Compared with the evaluation results of the single-classifier model, the stacking ensemble model has improved in the four indicators of accuracy, precision, recall and F1-score, which verifies the efficiency and effectiveness of the model proposed in this paper.