<![CDATA[Student dropout prediction is one of the most widely used Machine Learning applications to support academic decision-making. However, predictive model performance is often affected by imbalanced data distributions, where the number of non-dropout students significantly exceeds the number of dropout students. This condition may cause Machine Learning models to favor the majority class and reduce their ability to identify students at risk of dropping out. This study aims to analyze the effect of applying the Synthetic Minority Oversampling Technique (SMOTE) on decision bias in Machine Learning models for student dropout prediction. The dataset used in this research is the Predict Students Dropout and Academic Success dataset obtained from Kaggle, consisting of 3,630 records. Two classification algorithms were employed: Random Forest, and XGBoost. Model performance was evaluated using Accuracy, Precision, Recall, F1-Score, ROC-AUC, and False Negative Rate (FNR). The results demonstrate that SMOTE improves the detection of at-risk students. For Random Forest, recall increased from 0.8908 to 0.9120, while FNR decreased from 0.1092 to 0.0880. For XGBoost, FNR decreased from 0.0986 to 0.0915. These findings indicate that SMOTE effectively reduces decision bias caused by data imbalance and improves the reliability of Machine Learning models for academic early warning systems.]]>
Copyrights © 2026
