Research On Stock Price Forecasting Based On Support Vector Machine

For stock investors,stock price prediction is crucial.Many domestic and foreign scholars have studied and proposed a stock price forecasting model to help investors obtain excess returns.This article studies the law of stock price changes and finds that the key to stock price forecasting lies in two aspects: the judgment of the trend and the forecast of the price.First,the trend shows the change process of the stock price,judging the trend to clear the direction of the price change;then,make a precise prediction of the change of the price.In previous studies,the importance of stock movements was often overlooked,which in turn affected the accuracy of the forecasting model.Starting from these two aspects,this paper proposes stock time series reconstruction and hybrid nuclear CS-SVM price forecasting model.For the problem of trend,this paper proposes the reconstruction of the turning point sequence of the stock trend.The stock price always fluctuates.If the highs and lows of the trend can be judged,investors will be able to get the best return.After studying the theory of ChanZhongShuoChan investment,this paper analyzes the morphological characteristics of the K-line graph of the stock market,and proposes a method to judge the turning point of the trend by using a FenXing structure,and then reconstructs the sequence.For price forecasting problem,this paper constructs a hybrid nuclear CS-SVM prediction model.On the one hand,cuckoo algorithm(CS)optimization support vector machine(SVM)parameters are proposed.The cuckoo algorithm is a novel heuristic global search algorithm.The experimental comparison between CS-SVM and the other two SVM prediction models optimized by intelligent algorithm shows that the search efficiency of the cuckoo algorithm is higher and the optimization results are more stable.On the other hand,the radial basis kernel function and the polynomial kernel function are combined according to certain weights,and a mixed kernel function suitable for the reconstruction sequence is constructed by adjusting the weight coefficients.By comparing with the prediction result of the single kernel function,it is proved that the mixed kernel function has better learning ability and promotion ability.