Unveiling the dynamics of learning behaviors in learning K-12 math: An exploration of an ASSISTments Dataset

This study delves into the dynamics between diverse learning behaviors among K-12 students and their learning gains using a dataset of 508 students learning three math skills in ASSISTments. Employing K-means clustering based on students’ initial and final skill mastery alongside their engagement level, three distinct clusters emerged for each skill, revealing varying degrees of learning from ASSISTments. By analyzing decision tree classification models for each skill using affective labels such as boredom and frustration, we hypothesize that students within the same cluster of a skill may exhibit heterogeneous learning patterns that affect their subsequent learning of new skills. Further exploration demonstrates that students who transit between clusters when learning new skills differ significantly in their initial and final mastery of previously learned skills and their affective labels associated with those skills. Regression analysis underscores that students’ initial and final mastery of antecedent skills have some influence on their subsequent mastery of new skills. Unraveling the intricate relationship between student learning behaviors and the effectiveness of ASSISTments offers valuable insights into tailoring AI-enhanced educational tools, not only for learning the current skill but also for preparing for the future learning of new skills.