A Fast Least Squares Support Vector Machine Training Approach

A Fast Least Squares Support Vector Machine Training Approach (FTLSVM) to classification problem is proposed in this paper. The classification plane of FTLSVM is generated by solving a linear system of equations instead of a quadratic programming problem as for SVMs that is not fit for solving large-scale classification problems. Some simple techniques are used to solve the linear system to obtain fast computational time. The proximal support vector machines (PSVM) maximizes both direction w and threshold b to obtain faster computational time. In the paper our approach maximizes the margin between the two bounding planes with respect to the direction w. Our approach is based on LS-SVM, which gives results that are comparable to SVMs in use, in terms of test set correctness, but with considerably faster computational time. Lastly, the approach is compared with other approaches using synthetic and UCI datasets.