This study analyzed the trajectories of mathematical comprehension among 461 eighth-grade students using online learning log data, identified distinct learner types, and compared their characteristics across clusters. The data was collected from students who studied the topic “Linear Inequalities and Systems of Linear Equations” between April and June 2020, and Dynamic Time Warping (DTW) based K-means clustering was applied. The analysis yielded four learner clusters. Cluster 1 represented a Low-Level group characterized by consistently low achievement and short response times; Cluster 2, labeled the Stable High group, showed high achievement and sustained engagement; Cluster 3, the Gradually Declining Mid group, exhibited a gradual downward trend from a moderate level; and Cluster 4, the Rapidly Declining Lower-Mid group, demonstrated a sharp decline from a lower-mid level. Cluster 2 recorded the highest number of solved items, test-taking days, and response times, whereas Cluster 1 displayed features resembling skipping behavior. A higher proportion of male students was observed in Clusters 1 and 4, both of which also demonstrated the lowest achievement levels. Across all groups, the lowest accuracy rates appeared in Chapter 1 (Solutions to Inequalities) and 4 (Systems of Two Linear Equations in Two Variables), indicating that students experienced greater cognitive burden when acquiring new concepts. These findings provide evidence that online learning log data can precisely capture the relationship between learner behavior and achievement, and underscore the importance of identifying declining-achievement groups at an early stage to design targeted learning support strategies.