Predictive learning analytics for video-watching behavior in MOOCs

Madhumitha Shridharan, Ashley Willingham, Jonathan C. Spencer, Tsung-Yen Yang, Christopher G. Brinton

Abstract: In this paper, we develop Predictive Learning Analytics (PLA) methodology for learner video-watching behavior in Massive Open Online Courses (MOOCs). After defining features to summarize such behavior from clickstream measurements, we perform a statistical analysis of a real-world MOOC dataset and uncover several interesting relationships between the different features. Motivated by this analysis, we propose three algorithms for predicting future video-watching behavior, which incorporate biases for learners and videos, collaborative filtering across videos, and regularization to reduce overfitting. Through evaluation on our dataset, we find that the predictors obtain low RMSE overall, and that augmenting the bias predictor with either collaborative filtering or regularization improves prediction quality in eight out of nine cases.