Improved Weighted Learning Support Vector Machines (SVM) for High Accuracy

Support Vector machine (SVM) is a linear model designed for classification problem and popular due to a number of their attractive features such as high generalization ability and promising performance. However, the high generalization ability of SVM is only achieved by depending on a small part of the data points to determine an optimal hyperplane. During the learning process, the noise still exists to deviate severely the corresponding decision boundary from the ideal hyperplane. Two different weighted SVM such as one-step WSVM (OWSVM) and iteratively WSVM (iWSVM) has been reviewed besides the standard SVM. This method assigns relative important weights to achieve optimal margin hyperplane. In this study, an improved WSVM using moving weighted average is introduced to generate useful weighted and unweighted support vector for the optimal margin hyperplane. The methods are compared based on correctly labeled, mislabeled data within margin and classification accuracy using three datasets in KEEL repository with 20% noise. The results show that the proposed method yields better classification accuracy compared to OWSVM and iWSVM.