<![CDATA[Identifying students at risk of not completing STEM courses is essential for timely academic intervention, yet institutional learning data remain underutilized. This study develops a comparative learning analytics framework for classifying STEM course completion using Logistic Regression, a Shallow Artificial Neural Network (ANN), and a Deep Neural Network (DNN). The dataset contained 1,800 records from 300 students in the 2022 and 2023 cohorts, covering six STEM courses and 14 predictors. Data were stratified into training, validation, and testing subsets of 70%, 15%, and 15%, with random oversampling applied only to the training set. Performance was evaluated using accuracy, precision, recall, F1-score, AUC, and confusion matrices, with non-completion as the positive class. Logistic Regression achieved 97.78% accuracy, 94.57% precision, 98.86% recall, 96.67% F1-score, and 99.93% AUC. Shallow ANN achieved 97.41% accuracy and 96.09% F1-score, while DNN achieved 97.78% accuracy, 95.56% precision, and 96.63% F1-score. Final-exam, midterm, assignment, and quiz scores were the most influential predictors. These findings indicate that Logistic Regression provides the strongest overall balance between predictive performance and interpretability for STEM learning analytics.]]>
Copyrights © 2026
