Earliest predictor of dropout in MOOCs: a longitudinal study of FutureLearn courses

Abstract

Whilst a high dropout rate is a well-known problem in MOOCs, few studies take a data-driven approach to understand the reasons of such a phenomenon, and to thus be in the position to recommend and design possible adaptive solutions to alleviate it. In this study, we are particularly interested in finding a novel early detection mechanism of potential dropout, and thus be able to intervene at an as early time as possible. Additionally, unlike previous studies, we explore a light-weight approach, based on as little data as possible – since different MOOCs store different data on their users – and thus strive to create a truly generalisable method. Therefore, we focus here specifically on the generally available registration date and its relation to the course start date, via a comprehensive, larger than average, longitudinal study of several runs of all MOOC courses at the University of Warwick between 2014-1017, on the less explored European FutureLearn platform. We identify specific periods where different interventions are necessary, and propose, based on statistically significant results, specific pseudo-rules for adaptive feedback.