979-8-3503-3286-5/23/$31.00 ©2023 IEEE A Multi-modal Deep Learning Approach for Predicting Dhaka Stock Exchange Md. Nabil Rahman Khan 1 , Omor Al Tanim 2 , Most. Sadia Salsabil 3 , S.M. Raiyan Reza 4 , Khan Md Hasib 5 and Mohammad Shafiul Alam 6 1,2,3,4,6 Department of Computer Science and Engineering, Ahsanullah University of Science and Technology, Dhaka, Bangladesh 5 Department of Computer Science and Engineering, Bangladesh University of Business and Technology, Dhaka, Bangladesh *Corresponding Author: 190104044@aust.edu Abstract—This study proposes a reliable and accurate ap- proach for forecasting future stock price movements on the Dhaka Stock Exchange (DSE). Despite some people’s beliefs that it is difficult to create a predictive framework that can properly anticipate stock prices, there is a substantial body of literature that shows that seemingly random movement patterns in stock prices can be forcasted with a highly accurate result. The frame- work described in this study combines LSTM, Transformer, and the GRU model. Performance metrics including mean squared error (MSE) and R-squared (R2) are used to gauge the suggested DeepDse model’s accuracy. The results of the evaluation indicate that the model is highly accurate and can be used to provide reliable predictions of stock prices. This is of great importance, as accurate predictions of stock prices can assist investors in determining the best timing to buy and sell their investments. This can help investors minimize the risk of losing money and maximize their returns. The study suggests that the proposed model could be particularly valuable for investors in the Dhaka Stock Exchange, as it can provide them with valuable information to make informed investment decisions. Index Terms—Dhaka Stock Exchange (DSE), LSTM (Long Short-Term Memory), Transformer, Gated recurrent unit (GRUs), Time Series Data, Moving Average, Prediction I. I NTRODUCTION In today’s fast-paced economy, the stock market determines an organization’s and individual’s financial status. Predicting stock prices is vital since they may make or ruin a person’s fortune. Finance, statistics, and economic professionals have struggled with this problem. Due to their volatile, compli- cated, and ever-changing nature, stock price predictions have always been inaccurate. Analysts and traders find stock market prediction difficult. Stock markets are attractive and liquid investment options for corporations and individuals. Financial forecasting helps investors boost earnings. Many companies hire scientists and financial analysts to evaluate historical time series data to uncover patterns for better investment choices. Time series analysis uses statistical methods to identify pat- terns in this data and anticipate future trends [1]. Fundamental and technical analysis are the main stock market forecasting approaches. Fundamental analysis uses revenue, costs, and growth rates to anticipate stock values. Technical analysis analyzes previous stock prices and patterns to detect trends and forecast future prices. Fundamental re- search shows the company’s financial health, whereas techni- cal analysis shows market movements. Financial experts predicted stock market changes. Data and computer scientists are using data science and machine learning to increase prediction model performance and forecast accuracy. Deep learning, the latest level, improves prediction models. Data scientists struggle to build predictive algorithms for stock market predictions [2]. Complexity and nonlinearity arise from stock market volatility and investment psychology. Unpredictable elements like a company’s public image or a country’s politics can affect stock market trends. However, preprocessed share price data and the correct algorithms may predict stock prices and indexes. Machine learning and deep learning may help traders and investors make judgments by finding patterns in large volumes of data. Predicting price fluctuations using self-learning algorithms improves trading. Fig. 1 shows the basic structure of Recurrent Neural Network which represents the models we used in our paper. Fig. 1: A Recurrent Neural Network Artificial Neural Networks (ANN), Random Forest, LSTM, ARIMA, KNN, XGBoost, Gradient Boosting Machines (GBM), Time Series Analysis, Hidden Markov Models (HMM), SVMs, Naive Bayes and LightGBM were used to analyze historical market data and make predictions about future trends. These models are capable of automatically 0879 2023 IEEE 13th Annual Computing and Communication Workshop and Conference (CCWC) | 979-8-3503-3286-5/23/$31.00 ©2023 IEEE | DOI: 10.1109/CCWC57344.2023.10099255 Authorized licensed use limited to: Charles Darwin University. Downloaded on April 20,2023 at 06:12:12 UTC from IEEE Xplore. Restrictions apply.