Research On Classification Learning Machine Based On The Rescaled Hinge Loss Function

The classification problem is one of the important research problems in machine learning.It widely exists in speech recognition,text classification,disease diagnosis and other fields.In the process of data collection,outliers are often generated due to incorrect labels or noisy data.The occurrence of outliers will reduce the performance of the classi-fication model,and a robust classification method is needed to solve this problem.On the premise of enhancing the robustness of the model,this paper proposes a method to solve the binary classification problem and the multi-classification problem,as follows:First,a method to solve the binary classification problem is proposed,that is,twin support vector machine based on the rescaled hinge loss function.The main idea is:By combining the rescaled hinge loss function and the traditional twin support vector ma-chine,the optimization problem is obtained.In fact,the method is a weighted twin support vector machine,which gives different punishment to each sample point,and enhances the robustness of the model.This method can be applied to a wider range of problems,and it can solve the problem of cross-shaped data through a linear decision function.Finally,this method only needs to solve two small-scale quadratic programming problems,and the computational complexity is relatively low.Experiments based on artificial datasets and UCI datasets show that the algorithm has better robustnessSecond,a method to solve the multi-classification problem is proposed,that is,multi-ple birth support vector machine based on the rescaled hinge loss function.Its main idea is to combine the rescaled hinge loss function with the traditional multiple birth support vec-tor machine to obtain multiple birth support vector machine based on the rescaled hinge loss function.Firstly,it is actually obtained by giving different penalty points to each sample point,which can identify outliers in the dataset and reduce the impact of outliers on the model.Secondly,the weighted multiple birth support vector machine only needs to solve K small-scale quadratic programming problems,which has low computational complexity.Finally,the use of the rescaled hinge loss function turns the optimization problem into a non-convex optimization problem.In this paper,the conjugate function theory is used to equivalently transform the problem,and then it is solved using an alter-nating optimization iterative algorithm.Experiments based on noise-free and noisy UCI datasets show that the multiple birth support vector machine based on rescaled hinge loss function is effective.Compared with the existing multi-classification algorithm,it has higher classification accuracy.