<![CDATA[Academic failure and student dropout remain significant challenges in higher education because they affect educational quality and institutional performance. One of the major challenges in developing an effective dropout prediction model is class imbalance, which causes classification algorithms to be biased toward the majority class and reduces their ability to identify minority-class instances. This study aims to identify the dominant factors influencing academic failure by integrating the Hybrid Synthetic Minority Oversampling Technique–Edited Nearest Neighbour (SMOTE-ENN) with Ensemble Learning algorithms, including Decision Tree, Random Forest, XGBoost, and Voting Ensemble. Hybrid SMOTE-ENN was selected because it combines minority-class oversampling with noise and overlapping data removal, resulting in a more balanced training dataset and improving classification performance. The experiment was conducted using the Predict Students' Dropout and Academic Success dataset containing 4,424 student records. The research procedure consisted of data preprocessing, train–test splitting, Hybrid SMOTE-ENN resampling, model training, performance evaluation using accuracy, precision, recall, and F1-score, followed by feature importance analysis. Experimental results demonstrate that XGBoost with Hybrid SMOTE-ENN achieved the best performance, obtaining an accuracy of 87.12%, precision of 87.20%, recall of 87.12%, and F1-score of 87.15%. More importantly, the proposed model achieved a dropout-class recall of 80.99%, indicating its effectiveness in identifying students at risk of academic failure. Feature importance analysis revealed that Curricular Units 2nd Semester (Approved), Curricular Units 1st Semester (Approved), and Curricular Units 2nd Semester (Grade) are the three most influential factors affecting student dropout risk. These findings contribute to the development of an early warning system based on machine learning to support academic decision-making for imbalanced educational datasets.]]>
