Improving Lap-TSVM with Successive Overrelaxation and Differential Evolution

Recent proposed laplacian twin support vector machine (Lap-TSVM) gains good generation by solving a pair of quadratic programming problems (QPPs). However, the training procedure of Lap-TSVM is time-consuming. More- over, compared with SVM and TWSVM, Lap-TSVM has more parameters need to regulate, which affects its practical applications. In this paper, we improve the Lap-TSVM from the following two aspects: (1) By introducing the suc- cessive overrelaxation (SOR) technique, the QPPs of Lap-TSVM are solved with fast training speed without loss of generalization. (2) A differential evolution (DE)-based model for Lap-TSVM's parameters selection is further sug- gested. Our DE-based model uses the real-value encoding instead of binary numbers, which enhances the efficiency of parameters selection greatly. Computational results on several benchmark datasets confirm the merits of the greatly improvements on the training procedure of Lap-TSVM.