Early detection of students at risk of poor performance in Rwanda higher education using machine learning techniques

The prediction of student performance is one of the major issues in many higher education institutions throughout the world. This issue is lately detected due to the lack of a proper continuous monitoring framework. Failing to detect and assist students leads them to the uncertainty of success. Therefore, there is a need to develop an early detection framework to identify students who are at risk of poor performance for a timely intervention. This paper aimed at proposing a Machine Learning based framework for the early detection of students at risk of poor performance in Higher Learning Institutions in Rwanda. To achieve this objective, selected machine learning models have been trained and tested using the dataset of both secondary school leavers data and higher institutions data. Since the data was imbalanced, the Synthetic Minority Oversampling Technique method has been used to get a balanced data set. Based on the accuracy evaluation metric criterion, the Decision Tree model has been chosen with 63.18% of accuracy and the model has been used to compute the feature importance diagram. Findings revealed that the High School Program is the best predictor of poor performance followed by the Senior 6 aggregate. The findings will assist educational policymakers and practitioners in regularly monitoring students who are at high risk of poor performance and in considering possible mentorship to help them perform better.