The purpose of this study is to predict dropout learners in K-MOOCs and to cluster learners based on their learning characteristics, thereby providing strategic insights for the future operation of K-MOOCs. A total of 797 learners enrolled in four K-MOOC courses from March to May 2021 were analyzed. Learning activity and time-on-task data were used to conduct Random Forest and K-Means clustering analyses. The findings revealed several key predictors of dropout: average quiz scores below 48, fewer than 8 first quiz attempts, fewer than 8 correct answers, fewer than 16 quiz page views, time-on-task under 330 minutes, and dropout often occurring between weeks 3 and 4. Additionally, differences between completers and dropouts were more prominent in learning activity patterns than in timebasedmetrics. Cluster analysis identified three learner groups: Cluster 1 included learners with irregular patterns, some of whom completed the course despite inconsistent engagement, while others did not despite extended time and video usage; Cluster 2 consisted of learners with minimal engagement and short time-on-task who did not complete the course; Cluster 3 represented consistent learners who completed the course. Based on these patterns, adaptive instructional strategies tailored to each cluster were proposed to support more effective and personalized learning in online environments. 

Keywords : Social Learning Analytics, MOOC, Dropout Prediction, Cluster Analysis, Adaptive Learning, Online Learning Behavior ​