MODIFIED WEIGHTED SUPPORT VECTOR MACHINE (WSVM) ALGORITHM USING MULTIPLE HYPERPLANES AND INSTANCEWEIGHTED FOR CLASS NOISE

In this paper, a modified Weighted Support Vector Machine (WSVM) algorithm using Multiple Hyperplanes and Instance-weighted (MWSVM-MHI) will be proposed to solve class noise problem. The multiple hyperplanes is one of the approaches for selecting a subset of data that resembles the original data by reducing the number of samples in training data. A penalty term is added to the algorithm's cost function where it represents the size of the weights of the algorithm. In addition, the weighted classifier can deliver better classification performance than the unweighted classifier. The proposed method are evaluated in comparisons with WSVM and WKM-SVM based on three KEEL binary classification dataset repository namely Spambase, Sonar, Mines vs. Rocks, and Pima Indians Diabetes dataset. Each training data of the dataset will be corrupted by 20% class noise. It is shown that the proposed method produce higher classification accuracy in training data and test data for all datasets.