Feature selection methods for financial engineering

This paper explores the application of feature selection methods for financial engineering, and in particular the prediction of stock price movements. In the literature of feature selection methods, wrapper methods are found to be more accurate in generating the optimal subset. This is because the induction algorithm is used as a black box in the feature selection process. In this paper, a linear forward selection method is proposed as a subset search method to reduce the number of iterations and thus increase the efficiency of the wrapper. Stock data from 7 blue chip multi-national companies that are part of the Dow Jones Industrial Average (DJIA) from different sectors is used. The data used consists of over 5 years of data and 65 features. The class labels are generated based on the percentage change in closing price the following day. A wrapper method with linear feature selection is used to select the optimal feature set. Subsequently, 10-fold cross validation is performed to analyze the performance. We expect the wrapper method to perform traditional filter methods. To add credibility to this hypothesis, consistency based feature selection, correlation feature selection, fast correlation based feature selection are also performed on the data sets. A comparative analysis is drawn between the methods and the wrapper method using linear forward selection is found to generally out-perform the other methods.Bachelor of Engineerin