Predictive Course Placement: Utilizing an Artificial Intelligence-Based Approach for Recommendations Through the Application of Cosine Similarity and Multi-Model Architectures

Resumen

Accurately placing students in appropriate academic programs, such as determining whether they are better suited for a particular course, is fundamental to student success and retention. This study aims to construct a recommendation model that uses data mining and machine learning techniques to guide currently enrolled students in their program to make better selections. The research proposes a multi-model architecture that integrates a content-based cosine similarity matching component with three distinct machine learning classifiers, namely Logistic Regression, Random Forest, and XGBoost. In addition, the models were trained and evaluated using 10-fold cross-validation. Performance evaluations revealed that XGBoost was the most effective individual classifier, achieving a mean accuracy of 91.7%, an ROC-AUC of 97.5%, and a positive-class F1 score of 90.4%. The ensemble model also demonstrated strong predictive and recommendation performance, achieving 90.6% accuracy and 96.8% ROC-AUC. Logistic Regression achieved an accuracy of 88.7%, while the Random Forest model lagged with an accuracy of 83.8%. The findings demonstrate that XGBoost and Ensemble models are the most reliable and effective classifiers for predicting student course placement on this specific dataset. This hybrid recommendation engine successfully translates historical and input data into actionable, data-driven insights to help students make informed decisions about their academic careers.