ABSTRACT
In this work, recently proposed learning algorithm Vector-Valued Regularized Kernel Function Approximation (VVRKFA) and Single-hidden Layer Feedforward Neural Network (SLFN) are combined together to form a hybrid classifier called VVRKFA-SLFN. In VVRKFA method, multiclass data are mapped to the low dimensional label space through regression technique and classification is carried out in this space by measuring the Mahalanobis distances. On the other hand, SLFN has exceptionally high-speed and can achieve high accuracy on unknown samples. In the proposed hybrid learning system, VVRKFA is used to map the patterns from high dimensional feature space to a subspace and SLFN is trained on these low dimensional vectors to determine the class labels. The presented technique is verified experimentally both on benchmark and gene-microarray data sets to show the effectiveness of VVRKFA-SLFN classifier with enhanced correctness and testing time.
