Credit scoring plays an important role in the risk managements and it is a typical classification task in pattern recognition. The Least Squares Support Vector Machine (LSSVM) have provedn to be a powerful tool for the classification. However, it is hard for traditional parameter tuning methods such as cross validation and grid search to find the global optimal parameters for the LSSVM. To address this issue, in this paper, we propose a hybrid particle swarm optimization (PSO) and the LSSVM approach (PSO-LSSVM) for credit scoring. The PSO was first used to search for global optimal parameters for the LSSVM, then the LSSVM was applied for credit scoring. The experimental results on two real-word credit datasets show that the the proposed PSO-LSSVM outperforms the competing state-of-the-art methods in terms of accuracy, which demonstrates that the proposition of the PSO-LSSVM is promising for credit scoring.
展开▼
