Answers are grounded in this paper and cite their sources.
This paper proposes a weekly predictive modeling framework for forecasting students’ final performance in a university-level programming course. Unlike approaches that primarily focus on binary success/failure classification, the proposed framework uses regression models to provide fine-grained predictions of final grades across the semester, enabling a more detailed view of student progress and risk evolution over time. This work builds on a dataset previously used in one of our published studies, collected from 36 programming course sections across the Fall 2024 and Winter 2025 academic terms, and incorporates additional data from Fall 2025. From this dataset, a comprehensive feature set is engineered to represent students’ academic backgrounds, assessment outcomes, assessment durations, digital learning interactions, and engagement patterns. Machine learning (ML) models were trained on the combined Fall 2024 and Winter 2025 datasets using cross-validation and hyperparameter optimization, then evaluated on an independent Fall 2025 cohort to assess their generalization capability across academic terms. In addition, weekly forecasting experiments were conducted to assess model performance at different stages of the semester and determine how early reliable predictions of final performance can be achieved. The proposed framework provides instructors with a practical mechanism for continuously monitoring students’ expected final performance throughout the semester. By shifting the focus from coarse outcome classification to regression-based forecasting, the proposed framework offers a more actionable and informative approach for supporting early academic intervention in programming education.
No takes yet. Share an insight, caveat, or question.
Omri et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: