A Default Discrimination Method for Manufacturing Companies by Improved PSO-based LS-SVM

Loan default evaluation and discrimination is a complicated issue because of its nonlinearity and uncertainty. Least square support vector machine (LS-SVM) has been successfully employed to solve regression and time series problem. This paper proposes a novel PSO-LS-SVM model based on the improved PSO algorithm to optimize parameters of LS-SVM, which is a new improved form by synthesizing the exiting model of PSO. Some evaluation indices, which are reduced without information loss by a genetic algorithm, are used to train PSO-LS-SVM and discriminate between healthy and default testing samples. A case study based on financial data acquired from listed companies has been carried out. Result has shown that the proposed model has a distinct improvement in the aspect of accuracy rate as compared to PSO-SVM, LS-SVM, SVM and BP neural network.