A Study on Feature Selection for Trend Prediction of Stock Trading Price

The movement of price is influenced by many factors or features in stock market. It is a challenging work how to select these features and provide the relation between them and the movement of price. This paper applies two recursive feature elimination (RFE) methods SVM-RFE and RF-RFE to feature selection in the trend prediction of stock price, where SVM-RFE and RF-RFE are based on the famous support vector machine (SVM) and random forest (RF) techniques, respectively. Both the stability and classification performance for the subset of features selected are investigated. The experimental results on nine shares from Shanghai Stock Exchange in China show that both SVM and RF are effective for the trend prediction, and SVM performs better than RF. In addition, SVM seems to be unaffected by the correlation and redundant features and it turns better for the most shares when more features are used for modeling. Therefore, a suggestion for RFE in the trend prediction of stock price is that it may be unnecessary for SVM while it is needed for RF.