Development Of an Improved Machine Learning Model for Stock Exchange Prediction System Using Long Short-Term Memory (LSTM)

Abstract

The stock market serves as a fundamental driver and indicator of a nation’s
economy, providing opportunities for investors to trade and invest in shares of
companies and organizations listed on stock exchanges. This process aids in
enhancing the capital base of these companies, thereby yielding profits for market
investors. However, a significant challenge within the stock market is its inherent
volatility and uncertainty. The prices of various companies' stocks may fluctuate
unpredictably, resulting in potential investment losses. This dissertation proposes
a machine learning algorithm designed to predict stock market activities, aiming
to assist investors in making informed investment decisions. Specifically, we
introduce a Recurrent Neural Network model, the Long Short-Term Memory
(LSTM), utilizing the Structured System Analysis and Design Methodology
(SSADM). The model focuses on predicting the stock performance of Banks in
Nigeria, using a dataset sourced from the Nigerian Stock Exchange Group
repository, covering the period from 2000 to 2011. Implemented in Python within
Google Collaboratory, a cloud-based open-source machine learning development
environment, the model achieved an accuracy of 97.8% with a low error margin.
This high accuracy makes the model suitable for stock market prediction, thereby
mitigating investment risks and enhancing investor confidence.