GMM Weighted Pinball Loss Twin Support Vector Machine

General twin support vector machine with pinball loss function uses pinball loss function which produce noise insensitivity and resampling stability. However, Pin-GTSVM gives same weights to each sample points, including outliers, and outliers will disturb the generation of classification planes. To reduce this disturbance, we propose a weighted version of Pin-GTSVM called GMM weighted pinball twin support vector machine (Pin-GWTSVM). The weight matrix is introduced into the objective equation by Gaussian mixture model to reduce the interference of outliers, and additional regular terms are introduced to prevent overfitting based on the principle of structural risk minimization. Numerical experiments on several UCI benchmark datasets show that the proposed Pin-GWTSVM has better classification accuracy and generalization ability than Pin-SVM, TSVM and Pin-GTSVM.