← Back to Publications List

AI for Stock Market Forecasting in Emerging Economies

Students & Supervisors

Student Authors
Roy Bipro
Bachelor of Science in Computer Science & Engineering, FST
Supervisors
Md. Mortuza Ahmmed
Associate Professor, Faculty, FST

Abstract

Background: Capital markets in emerging economies typically exhibit volatility, low transparency, and enormous reaction to external influences, so accurate forecasting is a daunting task. Standard statistical models rarely examine financial data, as it is frequently nonlinear and extremely dynamic. However, Machine Learning (ML) and Artificial Intelligence (AI) models have been especially promising, inasmuch as such models are good at discerning fragile patterns and correlations among huge time-series databases that more conventional models overlook. Objective: The work attempts to find out if AI-based forecasting models can forecast fluctuations in share prices in emerging economies. The work employs historical data to validate if machine learning can support investors and decision-makers in more efficient decision-making in uncertain market environments. Methodology: The secondary data were gathered from Dhaka Stock Exchange (DSE), Yahoo Finance, and Kaggle, as historical time series of stock prices, trade volumes, and selected macroeconomic variables. The forecasting models applied include: Long Short-Term Memory (LSTM) networks, Support Vector Regression (SVR), Prophet, as well as the traditional Autoregressive Integrated Moving Average (ARIMA) model. The model performance was measured by Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), and Mean Absolute Error (MAE). Results: The relative performance of Bangladesh bank share prices revealed that ARIMA always produced highest performance, attaining lowest error margins (RMSE: 0.91, MAE: 0.64, MAPE: 1.17%) on test data. Though successful in representing long-run relations, LSTM could not handle abrupt price variations, resulting in bigger error margins. SVR produced moderate performances under steady-state conditions, while it was weak in performance under volatile conditions. Prophet was successful in representing seasonal patterns, although it underperformed in capturing abrupt market movements. Overall, it is observed that ARIMA is still the most stable and accurate solution in terms of forecasting in volatile emerging economies. Conclusion: The results indicate that Bangladesh stock price predictions can be genuinely enhanced by merging machine learning methods with traditional forecasting models. Though ARIMA appeared strongest under current conditions, the adaptability of LSTM and SVR suggests that hybrid models may have even higher predictive power. The blending of such models with visualization software that is interactive can further bolster predictions in emerging economies as well as real-world benefits among investors and planners of policy.

Keywords

Stock price forecasts Dhaka Stock Exchange Machine Learning LSTM ARIMA Financial prediction models

Publication Details

  • Type of Publication:
  • Conference Name: 7th International Conference on Integrated Sciences (ICIS) 2025
  • Date of Conference: 25/10/2025 - 25/10/2025
  • Venue: Eastern University Campus, Ashulia, Dhaka, Bangladesh
  • Organizer: International Open University, Eastern University
View on Publisher