Generative Data Augmentation: Enhancing Imbalanced Data Sets with Synthetic Samples Using Conditional GANs

Imbalanced datasets present a significant challenge in machine learning, particularly in tasks such as fraud detection, medical diagnosis, and sentiment analysis. This paper proposes a novel approach using Conditional Generative Adversarial Networks (CGANs) to generate synthetic samples that enhance the performance of models trained on imbalanced datasets. By conditioning the generation process on class labels, CGANs produce realistic and diverse samples of the underrepresented classes. The proposed method is empirically evaluated on various benchmark datasets, demonstrating superior performance in comparison to traditional oversampling methods like SMOTE (Synthetic Minority Over-sampling Technique). Our results show that CGANs significantly improve the predictive accuracy, precision, recall, and F1-score, leading to more balanced and reliable machine learning models.

• Conditional GANs
• Data augmentation
• Imbalanced data
• SMOTE
• Synthetic data