Institutional Data Analysis and Machine Learning Prediction of Student Performance

Abstract

Concern over student progression and retention is driving the systematic analysis of the vast amount of student data recorded in institutional data sets and in learning management systems (LMS) data sets. This is motivated by the belief that student analytics will help uncover patterns of behaviour and predict student performance, and thus facilitate the deployment of supportive educational interventions. This paper is concerned with the investigation of the institutional data of a Nigerian university, and the predictive impact of its attributes on student performance, measured in terms of cumulative grade point average (CGPA). The statistical analysis of the data set reveals that age and marital status have no significant relationship with final CGPA, whereas gender and pre-entry score have a weak relationship but are not good predictors. The application of four representative machine learning methods, namely linear regression, support vector regression, decision tree and random forests indicate that the third year CGPA is a good predictor of final year CGPA. A higher accuracy is achieved by using the aggregate value of the CGPAs of three previous years. Support vector regression has the best performance in predicting the final CGPA, whereas decision tree is the least performing model.