Genetic Dual Borderline SMOTE

With the widespread adoption of machine learning in recent times, numerous practical and theoretical studies have been conducted to enable machines to learn rare events. Making successful predictions for the minority class in imbalanced datasets has become increasingly crucial. We propose a new method, Genetic Dual Borderline SMOTE, to improve prediction accuracy for imbalanced datasets. The steps of the newly developed SMOTE method, along with its performance, have been compared with frequently used SMOTE, Borderline SMOTE, and K-Means SMOTE methods across eight datasets and four different machine learning algorithms. We used F-1 score of the minority class as the metric for performance evaluation and comparison. Various parameter combinations have been tested for each machine learning model and SMOTE method, and the parameters yielding the best F1 score for each model and SMOTE pair have been used. Our results show that the Genetic Dual Borderline SMOTE method outperforms other SMOTE methods, providing more successful outcomes.