Comparative study of financial distress prediction via op timized SVM

Abstract

This paper investigates the development and modeling problem for financial distress prediction via optimized support vector machine (SVM). Based on parameters optimization and model selection idea, the swarm intelligence algorithm of Particle Swarm Optimization (PSO)-SVM is proposed for financial distress predicting process with strong coupling and nonlinear characteristics through the principle component analysis (PCA). Furthermore, Logistic regression (LR) algorithm is induced to make a comparison with Least-Square support vector machine (LS-SVM) and PSO-SVM. The simulation results show that the presented algorithms could get the satisfied accuracy effectively, and by contrast, PSO-SVM shows a better learning ability and generalization in financial distress predicting process modeling, and could establish predictive model with better accessibility.