Addressing stock market time series trends and volatility using optimised DE-LSTM model

Accurate time series prediction is the most challenging trend for research communities in the machine learning era. Stock market data is the most dynamic and volatile time series data that holds the world economy. Recent studies proposed various core and hybrid machine learning models to get accurate stock forecasting. Existing work put forward long short-term memory (LSTM) to implement sequential time series data. In this paper, a new nonlinear hybrid model is proposed using customised LSTM and differential evolution (DE) algorithms. DE brings the optimisation of selection of parameters and provides stability between complexity and learning performance of the hybrid model. The paper explores the forecasting accuracy of the stock market trends and volatility using hybrid model DE-LSTM. The proposed hybrid model obtained significant improvement as MAE, RMSE and MAPE are 0.21167, 2.48198 and 2.68331 respectively, practiced for a diversified portfolio of Bombay Stock Exchange, India (BSE30).