PREDICTION OF STOCK MARKET RETURNS IN NIGERIA USING LONG-SHORT TERM MEMORY RECURRENT NEURAL NETWORK MODEL

ABSTRACT

Accurate forecasting of the stock market returns is a challenging task due to the volatile and nonlinear nature of the stock return values. Prediction of financial time series using different models in machine learning and deep learning have gained some degree of notable accuracy. The recurrent neural networks (RNN) have bases on deep learning with feedback loops. The cases of gradient vanishing and explosion are commonly associated with the traditional RNNs. The design of LSTM is intentional in eliminating the problems and this has made the LSTM to have become famous in the modelling of data with some complex sequence. The goal of this research work is to improve the accuracy of stock market forecasts using machine learning and classification algorithms of neural network LSTM. The historical stock prices data (2nd January, 2014 – 22nd September, 2021) of the selected company is subjected to the data mining and information extraction for the purpose of understanding the hidden patterns and forecast the behavior trend in the future times. The results reviewed that the proposed RNN-LSTM outperformed ANN and Fuzzy-GA models deployed for the same stock price movements datasets using MSE and RMSE as 0.1942 to 0.6889/0.7192, and 0.4990 to 0.8300/0.8481 respectively. The minimal MSE and RMSE obtained shows the efficacy of the LSTM model in the prediction of stock market returns