Enhance generalized learning vector quantization using unsupervised extreme learning machine and intelligent K-means clustering

Abstract

In this paper we proposed an enhancement of GLVQ classifier using USELM and IK-Means clustering. USELM is used to transform feature data into more separable form. The clustering method used to initiate codebook during training process. The proposed method has been tested using synthetic dataset and benchmark dataset. The proposed method has been compared to previous method and commonly used method. Experiment result shows that in over all dataset, the proposed method still has highest accuracy compared to others. Compared to GLVQ based classifier, the proposed method has better accuracy with margin 7.42%, 10.29%, 11.80%, and 8.11% for GLVQ, FNGLVQ, IK-Means-GLVQ, and USELM-GLVQ respectively. Compared to commonly used classifiers the proposed method has better accuracy with margin 1.94%, 2.93%, 11.61%, 31.37%, and 2.91% for MLP, Tree (J48), Linear-SVM, Sigmoid-SVM, and RBF-SVM respectively.