MOOCS DROPOUT PREDICTION BASED ON HIDDEN MARKOV MODEL

Massive Open Online Courses (MOOCs) have gained great attention in online education and attracted numerous learners in recent years. Nevertheless, MOOCs are limited by high dropout rate, which is perceived by the educator community as a prominent problem. Herein, a MOOCs dropout prediction approach based on the Hidden Markov Model (HMM) was developed to model sequences of learner behaviors over time in a MOOC. The HMM was constructed by sample balancing, feature discretization, feature selection and model training. Tests on the KDD Cup 2015 public dataset demonstrate that the accuracy of the HMM dropout prediction model yields was 87.5%, which is at least 5% higher than those of conventional models. The HMM dropout prediction model emphasizes the effects of key behavior features over time of a MOOC learner, resulting in effectively improved prediction accuracy.