Student Success Prediction: An Analysis of the Demand for a Transfer Learning Approach

Abstract

Student interactions with Learning Management Systems (LMS) generate logs, which are usually stored, allowing to recover each student activity. Analysis of these data with data mining and/or learning analytics techniques have been provided a better understanding of student behavior and teaching-learning processes. In this context, a number of studies have been reporting promising results in the task of predicting student performance, which allows proactive actions to avoid academic failures. Usually, data mining techniques estimate predictive models by using (past) historical data, assuming the premise that the estimated predictor will make predictions in future contexts that are similar to the (past) contexts which were used in its design. Although it is reasonable to assume that the diversity of existing educational contexts is reflected in the data, few studies discuss the impact of the aforementioned premise in the area of Educational Data Mining (EDM), resulting in models that may perform poorly when used under unforeseen educational conditions. This paper proposes an empirical analysis to verify evidences of differences between data from different educational contexts in the task of predicting students’ academic failure. Logs of more than 3,000 distance higher education students are used, and the adopted methodology is based on the supervised classification approach, commonly used in prediction tasks. Specifically, we aim to verify if distinct educational contexts are in fact separable in terms of the data they generate. Although data scenarios involve activities common to students in the same subject, the experiments indicate an accuracy of up to 83% in the separation of data from different academic terms. Although empirical, our results indicate a similar direction to that pointed out by other studies, contributing about the need of using transfer learning and/or domain adaptation techniques in the design of predictive models that aim to support proactive actions to prevent student failures.