How to Generate Early and Accurate Alerts of At-Risk of Failure Learners?

The high failure rate is a common serious issue among online learning institutions. In order to deal with this problem, Early Warning Systems (EWS) based on Machine Learning (ML) models are widely adopted in the literature as a solution to help teachers in their pedagogical monitoring. As the name implies, alert generation is intended to be the purpose of an EWS. However, the proposed systems don't go beyond the early prediction of at-risk of failure learners and don't suggest automatic methods to generate alerts. In this paper, we propose an algorithm that automatically generates early and accurate alerts for teachers of at-risk of failure learners. This algorithm uses both an original concept of alert rule to define the alerting method and temporal evaluation metrics to identify the reliable starting time for generating alerts. As a proof of concept, we apply this algorithm on four different EWS using real data of k-12 learners enrolled in online learning courses.