Support vector machine based multiagent ensemble learning for credit risk evaluation

The impact of the diversity of individual intelligent agents on the generalization performance of the SVM-based multiagent ensemble learning system is examined and analyzed.

Abstract

In this paper, a four-stage support vector machine (SVM) based multiagent ensemble learning approach is proposed for credit risk evaluation. In the first stage, the initial dataset is divided into two independent subsets: training subset (in-sample data) and testing subset (out-of-sample data) for training and verification purposes. In the second stage, different SVM learning paradigms with much dissimilarity are constructed as intelligent agents for credit risk evaluation. In the third stage, multiple individual SVM agents are trained using training subsets and the corresponding evaluation results are also obtained. In the final stage, all individual results produced by multiple SVM agents in the previous stage are aggregated into an ensemble result. In particular, the impact of the diversity of individual intelligent agents on the generalization performance of the SVM-based multiagent ensemble learning system is examined and analyzed. For illustration, one corporate credit card application approval dataset is used to verify the effectiveness of the SVM-based multiagent ensemble learning system.