Prediction of mathematics performance using educational data mining techniques

Higher Learning Institutions (HLIs) nowadays store a large amount of students' data. However, these data are not widely used to solve the students' academic problems at the institutions such as poor performance in some courses. Educational Data Mining (EDM) is a technology that can be applied to predict the performance of students from the dataset at HLIs. This study intended to solve the problem of poor performance in mathematics by management degree students at HLIs using EDM techniques and Mzumbe University (MU) in Morogoro, Tanzania as a case study. A quantitative research approach was applied based on the design science steps. Secondary data were collected to create the dataset through a review of documents from the examination, admission, accommodation, and accounts offices, as well as the Department of Mathematics and Statistics from the Main and Mbeya campuses of MU. Different Machine Learning (ML) algorithms were applied on the training set (60%) such as K-Nearest Neighbor (K-NN), Random Forest (RF), Decision Tree (DT), Support Vector Classification (SVC), and Multilayer Perceptron (MLP). Machine Learning algorithms were validated using a 10-fold cross-validation and validation dataset (20%) and the best algorithms were established to be RF, DT, and K-NN. Further evaluation of these three ML algorithms using 20% of the dataset demonstrated that the RF algorithm was the best for model development for the prediction of mathematics performance with an accuracy of 99% and F1-scores of 99% and 100% for the fail and pass classes respectively. Moreover, DT could generate rules that can be applied to recommend the minimum grade of D in ordinary level mathematics for admission into the University for Management Degrees to reduce the failure rates at HLIs.