Learning from Imbalanced Data Using Triplet Adversarial Samples

The imbalance of classes in real-world datasets poses a major challenge machine learning and classification, traditional synthetic data generation methods often fail to address this problem effectively. A limitation these is that they tend separate the process generating samples from training process, resulting lack necessary informative characteristics for proper model training. We present new method addresses issue by combining adversarial sample with triplet loss method. This approach focuses on increasing diversity minority class while preserving integrity decision boundary. Furthermore, we show reducing equivalent maximizing area under receiver operating characteristic curve specific conditions, providing theoretical basis effectiveness our In addition, further improve generalization small diverse set optimized using proposed function. evaluated several imbalanced benchmark tasks compared it state-of-the-art techniques, demonstrating can deliver even better performance, making an effective solution problem.