Stock Selection Algorithm Based On Support Vector Machine

Because stock forecasting is an uncertain, nonlinear and nonstationary time series probleam, it is diffcult to achieve a satisfying prediction effect by traditional methods. In this paper, a new prediction method based on Support Vector Machine (SVM) has been proposed. SVM can be used to solve many problems that traditional methods cannot solve effectively.The SVM training problem is described as a large-scale convex quadratic programming problem. The primary optimization problem and its dual problem are first driven for the case when all training data can be linearly separated. When the training data are separated by a nonlinear hypersurface, we transform the data to higher dimensional space such that the data will be linearly separable. First, this paper introduces the background knowledge of stock market, then traditional prediction methods are introduced in detail, and then the basic principles of SVM are discussed.Second, In order to improve the efficiency of the algorithm, this paper presents BFM-method to optimize the input vector of SVM. Result shows that this method can get similar result while using less computation time and less storage space.Third, this paper researchs the problem of parameters selection of kernel function, and compares serveral types of parameters of kernel functions by k-fold cross validation. Suitable parameters of kernel functions are chosen.At last, this paper compares the research results with the fund. The comparation shows that it is very suitable to use SVM to solve this problem. I believe that SVM will be an important method in the problem of predicting stock market.