Stock Price Classification Based on Hybrid Feature Selection Method

Abstract

In recent years, investors and traders have used Technical Indicators (TIs) to forecast the stock market. An accurate classification model is required in the stock market to gain more profit. Selecting relevant TIs for the stock market remains a hot research topic. The proposed work aims to identify important technical indicators. Therefore, the proposed work considers a hybrid feature selection method to identify the relevant TIs. The hybrid feature selection combines two individual feature selection methods, such as Boruta and the Random Forest (RF) feature importance method. The work considers 20 TIs. The regression power of TIs is computed using the hybrid feature selection method. Using the hybrid feature selection method, the selected relevant TIs are given as input to the classification model, namely Naive Bayes (NB) and Deep Learning. The classification model aims to classify the stock price as up or down. The Hybrid Feature selection-based Deep Learning H₂O model performs better than the hybrid feature selection-based NB model in the experimental work. The accuracy of the hybrid feature selection of the Deep Learning H₂O model is around 86 to 89%. The work considers the National Stock Exchange (NSE) in India for the experimental work.