The Research And Application Of PSO-SVM In The Stock Prediction

Stock occupies an important position in the market economy, the company issuing the shares can expand the scale by financing from the society,and the individual can affect the operation of the company by the way of holdings, it can be say that the stock has played an important role in promoting the development of market economy. Therefore, analyze and predict on the stock market is not only helpful for the individual to profit but also benefit for the decision-makers on macroeconomic adjustment, so that maintain the stable development of the national economy.With the developments of the statistical machine learning, many intelligent algorithms base on the statistical principles are emerging. From the view of the stock, it will produce great uncertainty in short-term investments, but in the long-term trend it conforms the statistical regular. Therefore, in the case of finite samples, using machine learning algorithms to predict the stock is an important research direction.The algorithm in this paper is base on the statistical machine learning principle after the research of the stock prediction techniques. At first,we analyze the feasibility of stock prediction using statistical machine learning principles, and then propose a new stock prediction method which regard the support vector machine classification as the core. In this method,we use the K-means clustering to classify the stock historical data from morphology, and then extract classic state indicators each type of historical data as the feature for predict.In the train period we train the data with support vector machine, during training use particle swarm optimization algorithm to adjust the key parameters of support vector machines,so that can improve the classification accuracy. In the predict period, we first classify the samples to be predicted by nearest neighbor classifier, and then use the related SVM to predict. The classification accuracy has been improved by this algorithm.In order to assess the effectiveness of the algorithm really and comprehensively, we use the SSE historical stock data as of the as the training set and prediction set to be analyzed. In the result of the experiment, including the compare of the algorithm proposed in this paper with other prediction algorithm. The experiment results show that the algorithm has a distinct progress in the prediction accuracy and adaptability, the predictive rate improved obviously.