Time-Series Clustering for Exploring Patterns of Online Learning Behavior Based on Dynamic Time Warp

In this study, we conducted a time-series clustering analysis based on dynamic time warping distance to understand learners’ behaviors observed in an online learning environment. We used log data from an English e-book learning system and compared the characteristics of learners in clusters derived from the analysis. Additionally, we compared the results of our analysis with those of aclustering analysis that did not reflect the time-series information of the log data. The main findings of this study are as follows: First, through a preprocessing process that focuses on the fluctuations in learning behavior observed over continuous time-series, we identified time-series patterns of learning behavior reflecting individual variability: one with high completion rates on weekdays, another with consistently low performance, and a third group exhibiting highly distributed patterns. Second, no significant differences in learning time and performance were found between the consistently low-performing group and the distributed group. Third, by comparing the clustering results with those obtained from the monthly average learning behaviors, the study highlighted the importance of this approach in terms of information retention. Based on these findings, the study discussed the need for alternative methods and methodological implications in processing and analyzing log data, where online learning behaviors are recorded in real-time.