Twin Bounded Large Margin Distribution Classifier Based on Unified Pinball Loss

Twin Support Vector Machine (TSVM) decomposes the optimization problem of the traditional Support Vector Machine (SVM) into two smaller optimization problems, which greatly improves the computing efficiency without reducing the classification performance. However, the Hinge Loss of the TSVM is only related to the shortest distance from the sample to the hyperplane, causing the hyperplane to be easily affected by noise and outliers. In order to solve this problem, this paper introduces the margin distribution measure of the distance from the sample to the hyperplane and the Unified Pinball Loss based on the TSVM, and proposes a Twin Bounded Large Margin Distribution Classifier based on the Unified Pinball Loss (UPTBLDC). The UP Loss is related to the quantile distance and is less affected by noise and outliers. In addition, in order to utilize the distribution characteristic information of the samples, the mean and variance optimization terms of the margin distribution are added to the objective function, which further improves the generalization ability of the model. Finally, a series of comparative experiments are conducted on the benchmark datasets and the datasets with artificially introduced noise, and the results showed that the proposed model has good noise immunity and stability.