Wrapper-Based Feature Selection and Optimization-Enabled Hybrid Deep Learning Framework for Stock Market Prediction

Abstract

Stock market is a significant element of economic market. Accurate forecasting of stock market is very helpful to shareholders because future prediction of a stock value will elevate the profits of investors. The data acquired from the stock market is a time-series data, in which the values of the stock prices are inherently varied with respect to time. Due to its complexity nature and nonlinearity characteristics, the prediction of stock market becomes very difficult and still it remains a challenging task. In order to cope up with such limitation, this research proposes an effective strategy called Deep Recurrent Rider LSTM to provide an accurate detection of stock market values. The accurate forecasting of stock market is carried out with two classifiers, namely Rider Deep Long Short-Term Memory (Rider Deep LSTM) and Deep Recurrent Neural Network (Deep RNN). The Rider Deep LSTM is derived by the integration of Rider concept with Deep LSTM, whereas the Deep RNN is trained using the proposed Shuffled Crow Search Optimization (SCSO). Moreover, the SCSO is derived by the integration of Shuffled Shepherd Optimization (SSO) algorithm and Crow Search Algorithm (CSA). Finally, the predicted output is determined based on the error condition. Furthermore, the proposed Deep Recurrent Rider LSTM achieved the MSE and RMSE of 0.018 and 0.132 that shows higher performance with better accuracy. The stock market prediction using the proposed classification model is accurate and improves the effectiveness of the classification.