Forecasting Dhaka stock exchange prices using machine learning models: a Performance analysis

Abstract

The financial markets have always been a focal point of interest for investors, analysts, and researchers. Predicting stock prices accurately has remained a challenging task. For many years, academics are analyzing the historical data to predict the prices; the most challenging and profitable use has been stock valuation forecasting. However, only a tiny portion of the elements which impact market movement can be measured. Examples of these factors include transaction volume, previous prices, and current prices. These variety of factors makes machine learning-based stock price prediction challenging and, to certain levels, questionable. Statistical and machine learning algorithms are used to predict short-term fluctuations in markets on an average market day, assuming there is ample historical data and factors available. This research uses a variety of machine learning techniques to present several comparison models for stock price prediction like LSTM, GRU and Nbeats and ARIMA. Historical data gathered from the official website of the Dhaka Stock Exchange (DSE) was used to train the models. Factors such as Date, Volume, Open, High, Low Close prices are included in the financial data. Conventional strategic metrics such as Mean Absolute Percentage Error (MAPE), Mean Squared Error (MSE) R-Squared and Mean Absolute Error (MAE) were used for assessing the models. Furthermore, since stock prices are impacted by various other real - world factors other than numerical data, this research attempts to incorporate external factors like political situations, daily grocery prices and corruption to the existing numerical variables in stock prediction. Our research contributes to the developing repository of knowledge in machine learning of machine learning for financial forecasting with significant implications for investors, financial institutions, and policymakers who depend on detailed stock price predictions to make informed decisions. The opportunity for more study in this area is outlined in the thesis conclusion, along with the practical ramifications of the results for the larger financial sector.