Research Of Stock Classification Based On An Improved Feature Selection Approach

Stock classification has been a challenging task and significant research field in finance and investment for both investors and researchers.The key to successful stock classification is achieving best results with minimum required input data and the least complex stock market model.Recently many researches have shown that a successful feature selection method can improve the classification accuracy of stock market.A key issue of the researches is how to select representative features for stock classification.This paper proposes a novel hybrid feature selection approach.This two stage method combines the advantages of three filter feature selection methods with an improved genetic algorithm(IGA)as a wrapper approach to identify an optimum feature subset and to increase the classification accuracy and scalability.Firstly feature ranking and feature weight was achieved by using 3 different filter algorithms.In the next step,the results of the filter stage were used as important prior information for the setting of initial population and improving genetic operators of genetic algorithm to speed up convergence.Finally,by bringing in multiple populations genetic algorithm(MPGA)which can enjoy the advantages of all filter feature selection algorithms and make a robust and accurate decision.The effectiveness of the proposed approach is evaluated in comparison with GA using A-shares data traded on the Shanghai Stock Exchanges from 2009 to 2013.The resulting database has 4173 data samples for 1108 companies and consists of 7 categories of financial ratios including 24 input features.The output variable is defined by the movement of stock return.The results demonstrate that the proposed hybrid feature selection method has the highest level of accuracy and better generalization performance than filter approaches and traditional genetic algorithm.Finally,the important features in stock classification are selected then stock return re-classed.The results show that the proposed hybrid model is a proper tool for effective feature selection and these features are good indicators for the stock classification.