Abstract:

The least square support vector machine only needs to solve a linear system of equations to get a closed solution,so it is widely used in classification problems because of its fast training speed.However,the least squares support vector machine model is easily affected by outliers and noise,which often reduces the classification accuracy.Fuzzy weighting of sample points is an effective method to solve this problem.Intuitionistic fuzzy sets contain both membership information and non-membership information of sample points,which can describe the distribution characteristics of sample points in more detail.Therefore,based on the intuitionistic fuzzy set,this paper obtains a more accurate class center by eliminating outliers,and then calculates the distance between the sample point and the class center to obtain the membership degree of the sample point to its class.At the same time,the kernel k-nearest neighbor method is used to find the number of k neighboring sample points of another class,and then the non-membership information of sample points is obtained.Finally,a new fuzzy value is obtained according to the membership degree and non-membership degree of sample points.Furthermore,the proposed fuzzy values are used to improve the LSSVM model.By assigning outliers and fuzzy values with low noise,their influence on the LSSVM model is reduced and the accuracy of the LSSVM model is improved.Experimental results show that,compared with the existing algorithms,the proposed algorithm can reduce the influence of outliers and noise on the LSSVM model and improve the robustness of the model.

Key words: least square support vector machine, fuzzy sets, outliers and noise, k-nearest neighbor