A framework for predicting preservice teacher performance using machine learning and SMOTE-based class imbalance handling

Abstract

Improving preservice teachers’ performance is crucial in higher education. By predicting their academic success, educators can identify at-risk preservice teachers, provide additional supports, and enhance their achievements. However, only a few studies have been undertaken to develop a predictive model focusing on preservice teachers’ performance. Therefore, this study was conducted to determine the most effective model for predicting preservice teachers’ academic outcomes by evaluating multiple machine learning techniques, including Decision Tree (DT), Random Forest (RF), Logistic Regression (LR), Naïve Bayes (NB), Support Vector Machine (SVM), K-Nearest Neighbor (KNN), and Artificial Neural Network (ANN) across five distinct groups. The models were trained and evaluated using fivefold cross-validation, with performance assessed by accuracy, precision, recall, and F1-score in individual groups. To examine the models’ performance tendencies and robustness across all samples, the optimal model of each sample was further compared through model occurrence frequency, average accuracy score, and average F1-score. The results highlighted disparities across samples, with no single algorithm systematically exceeding the performance of the other models, and showed no statistically significant difference between the models. However, SVM and KNN may serve as baseline predictive models in small-sample educational research and underscore the necessity of context-driven machine learning selection for predictive tasks. Overall, this study highlights the potential of machine learning algorithms in advancing teacher education and provides additional insights on research methodology and context-dependent algorithm selection.