Predicting student retention: A comparative study of machine learning approach utilizing sociodemographic and academic factors

Student dropout remains a persistent challenge in higher education, particularly in developing regions where institutional resources for intervention are limited. This study explores the predictive potential of machine learning (ML) algorithms in identifying students at risk of dropping out using historical academic and sociodemographic data from Mindanao State University–Main Campus, covering a ten-year period (2012–2022). A total of 482 student records and 146 variables were preprocessed using Power BI and prepared via the CRISP-DM methodology before being modeled in Jupyter Notebook. Ten ML algorithms such asExtreme Gradient Boosting (XGBoost), Gradient Boosting (GB), Artificial Neural Network (ANN), Decision Tree (DT), Random Forest (RF), Multilayer Perceptron (MLP), Logistic Regression (LR), K-Nearest Neighbor (KNN), Support Vector Machine (SVM), and Naïve Bayes (NB) were evaluated using both train-test split and 5-fold cross-validation to ensure robustness and generalizability. Results indicate that XGBoost outperformed all other models, achieving the highest cross-validated accuracy (90.66 %), F1 Score (90.72), and one of the lowest error values (Mean Square Error (MSE) = 9.34, Log Loss = 0.26). GB and ANN followed closely, demonstrating strong balance between precision, recall, and low misclassification rates. While models like Naïve Bayes recorded high recall, they also produced excessive false positives, limiting their practical use. The study offers a scalable and transferable modeling framework for higher education institutions seeking to implement early warning systems. It also emphasizes the pedagogical value of integrating educational data science into Information Technology Education (ITE) curricula to foster real-world AI application. Limitations and future directions are discussed, particularly regarding behavioral data integration and model interpretability.