Enhancing Classification of Imbalanced Data using CA-SMOTE

Abstract Imbalanced data classification poses challenges across domains like medical diag- nosis and fraud detection. While SMOTE addresses this by oversampling, its limitations include noise and overfitting. Variants like ADASYN offer improve- ments but still have drawbacks. To tackle these issues, CA-SMOTE emerges, leveraging both minority and majority classes to create diverse synthetic sam- ples. By closely aligning with the data distribution, CA-SMOTE aims to enhance model accuracy and reliability. Comparative evaluations demonstrate its supe- riority over SMOTE and other modern techniques in terms of F1-score and AUC-ROC. CA-SMOTE presents a promising avenue for improving imbalanced data classification across various real-world applications.