Optimizing Random Forest Classification Using Chi-Square and SMOTE-ENN on Student Drop-Out Data

Dropout is a particular concern for countries striving to increase human capital. Various attempts have been made by universities to minimize the number of dropouts. Machine learning has also developed various predictive models to determine the likelihood of students dropping out. However, there is a challenge in dropout data, specifically the problem of class imbalance, where the number of students who drop out (minority class) is significantly less than those who do not drop out (majority class). This imbalance can reduce the model’s ability to classify students at risk of dropping out. This study proposes classification optimization using the Random Forest algorithm to handle class imbalances in student dropout data. To overcome class imbalance, the Synthetic Minority Oversampling Technique (SMOTE) and Edited Nearest Neighbors (ENN) techniques are used. Additionally, the attribute selection method is also applied to enhance the predictive results. The test results demonstrate that the combination of implementing feature selection with Chi-Square, followed by class imbalance handling with SMOTE-ENN, provided the most optimal predictive performance for identifying the status of both dropouts and graduates.