PREDICTING STUDENT ACADEMIC PERFORMANCE USING MACHINE LEARNING: A CASE STUDY ON EDUCATIONAL DATA FROM BANGLADESH TOWARDS SUSTAINABLE EDUCATION (SDG 4)

Educational sustainability is a global priority. Because of the rapid digital transformation that we are seeing, it is necessary to integrate modern technologies to achieve inclusive, equitable, and high-quality education. Despite this, there is still a limitation in applying Machine Learning (ML), as one of the modern technologies, to educational data within institutions. This research addresses this gap by providing practical evidence that predicting academic performance via ML is both achievable and highly effective. Two main objectives were generated from the main goal. The first one is to compare four ML models, while the second is to determine the most important factors affecting academic performance. To achieve the previous goals, the study used data from more than 8,000 secondary school students from Bangladesh. Based on the given dataset, students’ academic performance was represented by the average grades of five common courses, where the study tried to predict it without relying on the existence of previous grades. The study depends on the contextual, behavioral, and other variables that are available in that dataset. Three of the models used were shallow: Multiple Linear Regression (MLR), Decision Tree (DT), Random Forest (RF), while one was deep: Feedforward Deep Neural Network (DNN). The whole process was implemented in the R environment, and the results revealed some interesting points. The shallow RF model was more accurate than the deep one with a small margin (R² = 93.3 vs. R² = 93.0). The most important factor affecting student performance is the student group (arts, commerce, or science), which dominated the predictive power of the models. Study time and attendance can also not be ignored as important behavioral modifiable factors. In addition to the results presented, this study contributes to educational sustainability by developing two Early Warning Systems (EWS): a simplified system for quick group screening and a full EWS for individual predictions. The results and the contributions of this study will help to find the most at-risk subgroups, particularly in the arts track, and allocate resources to provide proactive support based on modifiable behaviors or other factors. Ultimately, this research provides a scalable framework to enhance equity and efficiency in education. It is aligned with UN Sustainable Development Goal 4 (SDG 4): Quality Education.