Predicting Student Dropout based on Machine Learning and Deep Learning: A Systematic Review

Abstract

Student dropout is one of the most complex challenges facing the education system worldwide. In order to evaluate the success of Machine Learning and Deep Learning algorithms in predicting student dropout, a systematic review was conducted. The search was carried out in several electronic bibliographic databases, including Scopus, IEEE, and Web of Science, covering up to June 2023, having 246 articles as search reports. Exclusion criteria, such as review articles, editorials, letters, and comments, were established. The final review included 23 studies in which performance metrics such as accuracy/precision, sensitivity/recall, specificity, and area under the curve (AUC) were evaluated. In addition, aspects related to study modality, training, testing strategy, cross-validation, and confounding matrix were considered. The review results revealed that the most used Machine Learning algorithm was Random Forest, present in 21.73% of the studies; this algorithm obtained an accuracy of 99% in the prediction of student dropout, higher than all the algorithms used in the total number of studies reviewed.