Predicting early dropout students is a matter of checking completed quizzes: the case of an online statistics module

Abstract

Higher education students who either do not complete the subjects they enrolled in or interrupt indefinitely their studies without certification, the socalled college dropout problem, still continues to be a major concern for practitioners and researchers. Within the subjects, an early prediction of dropout students has aided teachers to focus their intervention in order to reduce dropout rates. Several machine-learning techniques have been used to classify/predict dropout students, including the tree-based methods which are not the best performers, but in their favour, are easily interpretable. This study presents a procedure to identify dropout-prone students at an early stage in an online statistics module, based on decision tree models. Although the attributes initially considered in the creation of the trees were mainly related to quiz completion, participation in the forum and access to the bulletin board, the final models show that the former is the only attribute with significant discriminatory power. We have evaluated the classification performance by means of a validation set. The performance measure of accuracy shows values above 90%, whereas that of recall and precision slightly under 90%.