An Intelligent Credit Assessment System by Kernel Locality Preserving Projections and Manifold - Regularized SVM Models

Support vector machines (SVM) have been successfully applied in numerous areas of pattern recognitions, and have demonstrated excellent performance. However, traditional SVM does not make efficient use of both labeled training data and unlabeled testing data. Moreover, one usually encounters high dimensional and nonlinear distributed data in classification problems, especially in financial credit rating assessments. They generally degrade the performance of a classifier due to the curse of dimensionality. This study addresses these problems by proposing a novel intelligent system which integrates a kernel locality preserving projection (KLPP) with a data-dependent manifold-regularized SVM. KLPP is employed to gain a perfect approximation of data manifold and simultaneously preserve local within-class geometric structures according to prior class-label information. Empirical results indicate that, compared with other dimensionality reduction methods and conventional classifiers, the hybrid classifier performs best.