Dropout Prediction in MOOCs using Learners’ Study Habits Features

Many educators have been alarmed by the high dropout rates in MOOC. There are various factors, such as lack of satisfaction or attribution, may lead learners to drop out. Educational interventions targeting such risk may help reduce dropout rates. The primary task of intervention design requires the ability to predict dropouts accurately and early enough to deliver timely intervention. In this paper, we present a dropout predictor that uses student activity features and then we add learners’ study habits features to improve the accuracy. Our models achieved an average AUC (receiver operating characteristic area-under-the-curve) as high as 0.838 (if lacking study habits is 0.795) when predicting one week in advance. The model with learners’ study habits features attained average increase in AUC of 0.03, 0.06, 0.08 and 0.05 in different cohorts (passive collaborator, wiki contributor, forum contributor, and fully collaborative).