Long Short-Term Memory and Discrete Wavelet Transform based Univariate Stock Market Prediction Model

Abstract

Analyzing financial situations in the current scenario is difficult, as it requires understanding the quality and value of investments. This study predicted the movement of stock prices in the Saudi Arabian stock market (Tadawul) over a one-week period using a proposed integrated model of Long Short-Term Memory (LSTM), which combines LSTM, Discrete Wavelet Transform (DWT), and Autoregressive Integrated Moving Average (ARIMA). Historical closing prices of a group of four companies listed on Tadawul were used as input for the proposed LSTM model, which consists of memory units capable of storing long time periods. Once the LSTM model predicted the closing values of stocks in Tadawul, they were further analyzed using the ARIMA model. The prediction accuracy of the proposed LSTM model and the traditional ARIMA model were 97.54% and 96.29% respectively. Therefore, the proposed integrated model of LSTM is considered a useful tool for predicting stock market values. The results emphasize the significance of Deep Learning (DL) and leveraging multiple information sources in predicting stock prices.