Applications of Educational Data Mining in Identifying Learning Difficulties among Novice Programmers

Abstract

This narrative review examines how educational data mining (EDM) can inform instructional decisions about learning difficulties among novice programmers. This paper draws on a focused set of publications, including empirical and application-oriented studies. The analysis focuses on target alignment, timing, and granularity. Submission histories and compiler feedback can reveal learning processes that are not reflected in final grades. Repeated attempts may indicate either progress or ongoing difficulties. Grouping learners and sequence analysis help identify behavioral patterns. Supervised prediction is used to estimate outcomes, while cognitive diagnosis assesses knowledge mastery. Early records may help prioritize students for review. Concept-focused support requires more detailed evidence and checks against actual assignments. Differences in targets, observation windows, and validation designs limit comparisons across studies. Results based on complete learning histories do not establish early prediction capability. Model explanations also do not establish intervention effects. The proposed framework remains empirically unvalidated. Evaluation should address measurement quality, generalizability, predictive error, privacy, and fairness. The effects of subsequent instructional support on learning require separate assessment.