Time Series Clustering of Learner Engagement in Large-Scale Online Learning Log Data Using Dynamic Time Warping

This study aims to analyze learners’ engagement patterns over time by applying time series clustering based on Dynamic Time Warping(DTW), using large-scale log data collected from a TOEIC learning platform. Indicators representing learning engagement-such as the number of items solved, lectures attended, and login frequency-were extracted and clustered using the k-medoids algorithm. The main findings of this study are as follows. The number of items solved and lectures attended were grouped into three clusters: learners with high activity in the early, middle, and late stages of the learning period. In contrast, login frequency was divided into two clusters, reflecting high activity either in the early or late stages. In this process, the DTW-based time series clustering method revealed that learners with similar learning patterns were grouped together even when their time series lengths differed. However, middle-focused learners, characterized by an increase followed by a decline in activity, were clustered together regardless of the specific timing of the change, indicating a limitation of DTW-based clustering. Based on these findings, this study discusses both the potential and the limitations of DTW-based clustering, and derives implications for analyzing large-scale learning log data and designing online learning systems.