This study addresses challenges in elective course selection among Information Technology students at KMUTT. Misalignment between chosen electives and decrease the risk of delayed graduation. we compare model-based and memory-based collaborative filtering approaches to identify an effective method for course recommendation . Using students’ academic histories, including their previously completed courses and grades, we evaluate each model’s accuracy and relevance in predicting suitable electives Our findings inform decision-support tools that assist students in making informed elective course selection Experimental results demonstrate that SVD achieves a strong balance across all evaluation metrics, attaining a QWK of 0.7029, an RMSE of 0.5567, and an adjusted accuracy of 82.63%, while maintaining a notably efficient training time of 0.1241 seconds the lowest among all evaluated models. Although SVD++ yields marginally superior predictive performance under certain parameter configurations, its substantially higher computational cost renders it less practical for deployment in real-world academic advising systems. Consequently, SVD is recommended as the most suitable model, offering a compelling trade off between predictive performance and computational efficiency based on their past academic performance helping them choose appropriate course. To ensure a rigorous evaluation that is highly suitable for ordinal grade data, we utilize the Quadratic Weighted Kappa (QWK) metric.

Umaporn Supasitthimethee

Niwan Wattanakitrungroj