Enhanced Prediction of Intra-day Stock Market Using Metaheuristic Optimization on RNN–LSTM Network

Abstract

Deep Learning provides useful insights by analyzing information especially in the field of finance with advanced computing technology. Although, RNN–LSTM network with the advantage of sequential learning has achieved great success in the past for time series prediction. Conversely, developing and selecting the best computational optimized RNN–LSTM network for intra-day stock market forecasting is a real challenging task as a researcher. Since it analyses the most volatile data, requires to cope with two big factors such as time lag and the large number of architectural hyperparameters that affect the learning of the model. Furthermore, in addition to the design of this network, several former studies use trial and error based heuristic to estimate these factors which may not guarantee the most optimal network. This paper defines the solution to solve the above-mentioned challenging problems using the hybrid mechanism of the RNN–LSTM network integrating with a metaheuristic optimization technique. For this, a two-hybrid approach namely RNN–LSTM with flower pollination algorithm and RNN–LSTM with particle swarm optimization has been introduced to develop an optimal deep learning model to enhance the intra-day stock market prediction. This model suggests a systematic method which helps us with an automatic generation of optimized network. As the obtained network with tuned hyper parametric values-led towards a more precise learning process with the minimized error rate and accuracy enhancement. In addition, the comparative results evaluated over six different stock exchange datasets reflect the efficacy of an optimized RNN–LSTM network by attaining maximum forecasting accuracy approximately increment of 4–6% using the metaheuristic approach.