Predicting Student Dropout in E-Learning Using Simple Machine Learning and Explainable Data Analysis

Abstract

Online learning allows a lot of flexibility and accessibility, but the issue of student dropout becomes one of the key problems. Much of the research that has been done to predict dropout using complex machine learning methods has not been interpretable, and many of these methods are hard to implement in practice, especially in resource constrained environments. This research closes this gap by presenting a simple and explainable machine learning method of predicting early dropout. Students in several universities were surveyed on the frequency of attendance, quiz performance, completion of assignments, satisfaction with learning, and hours spent studying per week. Because there were only a few real dropout cases, a controlled synthetic data augmentation method was used to demonstrate and train the model. The use of Logistic Regression and Decision Tree classifiers was used to predict the risk of dropout. The accuracy of both models was 87.5% with the most effective indicators being learning satisfaction, attendance, and prior consideration of dropout. This research is considered new because it focuses on the simplicity and interpretability of models rather than the complexity of creating early warning systems to predict the dropout of students, showing that they are not required to be complex or made out of heavyweight models to perform well. Even though the findings are merely exploratory due to limitations of the datasets, the results show that even simplistic models might be used to assist educators in identifying at-risk learners and incorporating prompt intervention measures.