Research On Prediction Of Stock Price Using Parallel Support Vector Machine Based On GPU

As an important part of financial market,stock market has the characteristic of nonlinear fluctuation.Therefore,how to predict the fluctuation of stock market price accurately has always been the focus of research.Many scholars at home and abroad have put forward some effective methods of stock price prediction,such as fundamental analysis,technical analysis and so on,but the forecasting effect is not satisfactory.With the development of nonlinear technology,artificial Neural Network(Ann)has been widely used in stock price forecasting,and has achieved good prediction results.However,ANN has many shortcomings based on the principle of empirical risk minimization.Such as slow convergence rate and easy to fall into the local minimum and so on.Support vector machine(SVM)based on structural risk minimization(SVM)can solve these problems well,so many scholars apply it to stock market.It is verified that support vector machine can predict the trend of stock market well.However,when dealing with a large number of data sets,SVM will show some shortcomings such as slow training speed,and with the increasing of data scale and dimension of stock market,the computational efficiency of SVM is not ideal.In recent years,GPU graphic Processing Unit has been applied in the field of general computing by virtue of its powerful floating-point computing capability,parallel computing capability and high storage broadband.Therefore,support vector machine(SVM)is parallelized by GPU technology and applied to stock price prediction.The specific work is as follows:(1)For the problem that the training speed of the support vector machine is too slow,based on the Sequential Minimal Optimization(SMO)algorithm,the parallelism is analyzed and a SMO algorithm based on the GPU is proposed.(2)Based on the SMO algorithm based on GPU and the mesh search algorithm and particle swarm optimization algorithm for parameter optimization,a parallel SVM stock price prediction model is constructed.(3)The parallel SVM stock price prediction model is tested by using four groups of stock data.The GS-SVM model,GPU-GS-SVM model,PSO-SVM model and GPU-PSO-SVM model are analyzed and compared in terms of the running time and prediction accuracy of the model.The experimental results show that the parallel SVM stock price prediction model optimized by the grid search algorithm based on GPU can significantly shorten the running time with better accuracy of stock price prediction.