SHU YA XU
Student Graduate School of Information Management Tunghai University Taichung City, Taiwan
E-mail：G14490003@thu.edu.tw
HSIN CHUN YU
Professor Graduate School of Information Management Tunghai University Taichung City, Taiwan
E-mail：hsyu@thu.edu.tw
Abstract
This study investigates the structural characteristics of student learning behaviors in online learning environments within the field of learning analytics, moving beyond outcome-oriented dropout prediction. Mapper-based topological data analysis is used to describe branching and continuity in high-dimensional activity traces. Using the Open University Learning Analytics Dataset (CCC-2014J), each learner is encoded as a “week × activity type” interaction vector. UMAP serves as the lens; a multi-scale cover with local clustering yields a behavioral graph, and role stability is examined across cover resolutions. The map indicates pronounced heterogeneity and four recurrent roles: early disengagement, assessment-driven but short-lived participation, sustained engagement, and isolated activity. Dropout follows multiple transitions, with intermediate nodes marking behavioral shifts over time. This structural view complements conventional models and supports the development of structure-aware features for explainable learning analytics.
Keywords：Topological Data Analysis, Mapper Algorithm, Online Learning Behaviors, Learning Analytics, Student Retention and Dropout