Application of Machine Learning Techniques to the Prediction of Student Success









Abstract

This study presents six machine learning models in the prediction of student success in a technology-mediated environment. Student behavioral attributes with a learning management environment have proven to be a significant determinant in forecasting students' performance. This study attempts to provide the model with optimum accuracy to determine students who need assistance to improve their educational performances and other learning outcomes. We examined the impacts of SMOTE data re-sampling and the effect of attribute selection in this study. The models' performances were enhanced with the resampling method as the imbalanced dataset was identified to have performed poorly. Attribute Selection with the top ten attributes and 10-fold cross-validation offer best performances. The six predictive models utilized in this study are Linear Discriminant Analysis, Logistic Regression, Classification and Regression Tree, K-Nearest Neighbour, Naive Bayes Classifier, and Support Vector Machines. Classification and Regression Tree model and Linear Regression had the best accuracy score of 0.86 after 10-fold cross-validation and top ten attribute selection. This study concludes that student behavioral attributes are useful predictors of student success.


Modules


Algorithms


Software And Hardware