Research On The Improvement Of Semi-supervised Kernel Clustering Algorithm Based On Multi-factor Stock Selection

This paper compares the shortcomings of traditional K-means clustering,semi-supervised K-means clustering,semi-supervised K-means kernel clustering for solving problems of model local optimal solution,non-linear distribution and non-spherical cluster of sample distribution in multiple-factor stock selection problem,and finally proposes an improved semi supervised K-means kernel clustering method.Based on the method of "semi-supervised K-means kernel clustering based on the influence factor of gravitation",the method has the advantages in dealing with the problems of the non-linear distribution and non-spherical cluster of sample distribution by improving the kernel function.The empirical results show that the Stock sample matrix is a typical high sparse cluster matrix,and the distribution of samples has obvious linear irreducible problems.The improved model has strong generalization ability and has obvious advantages in dealing with high sparse matrices and sample linear irreducible problems.The innovation of this paper is: 1.This paper combines the theory of kernel function with the method of "semi-supervised clustering based on the influence factor of gravity" to get the semi-supervised kernel clustering algorithm based on the gravity influence factor,and the new model has good effect on the problem of linear non-splitting of samples,2.On this basis,the kernel function is improved to obtain the "improved semi-supervised kernel clustering algorithm",and the improved model has better learning ability and generalization ability than the former model,3.This paper applies the model to the multiple-factor stock selection,and the empirical results show that the model can choose a better stock combination.