Prediction of student academic performance through Machine Learning

Abstract

The objective of this research study was to evaluate two machine learning techniques, including Decision Trees and Random Forests, in order to predict students' academic performance. The results indicated that the Random Forests algorithm with hyperparameters of 40 trees and a maximum depth of 20 levels, a higher accuracy of 0.77 is achieved. In addition, the value of the AUROC indicator in the ROC curve of the Random Forests algorithm is greater than the threshold of 0.7 reaching an acceptable estimate compared to the Decision Tree algorithm, whose value is 0. 69 thus demonstrating that the Random Forests algorithm was the most accurate in predicting academic performance, this finding is relevant for educational institutions in general, as it can help to define follow-up and support policies for those students at academic risk. In addition, the potential application of these machine learning techniques in the different careers of the Faculty of Industrial Process Engineering of the National University of Juliaca was estimated.