Robust Semi-Supervised SVM on Kernel Partial Least Discriminant Space for High Dimensional Data Mining

Kernel machines (such as support vector machines) have demonstrated excellent performance in numerous areas of pattern recognitions. However, traditional kernel machines do not make efficient use of both labeled training data and unlabeled testing data. Moreover, high dimensional and nonlinear distributed data generally degrade the performance of kernel classifiers due to the curse of dimensionality. To address these problems, this study proposes a novel hybrid classifier which constructs a robust semi- supervised support vector machine (SVM) on kernel partial least square discriminant space (KPLSDS). KPLSDS is created by optimal projection of original data space to a representative low dimensional subspace which has maximum covariance between inputs and outputs. Robust semi-supervised SVMs on KPLSDS exploit the candidate low-density separators and simultaneously prevent identifying a poor separator from the help of unlabeled data. Compared with other dimensionality reduction methods and conventional classifiers, the hybrid classifier performs best.