CEGAN: Classification Enhancement Generative Adversarial Networks for unraveling data imbalance problems. (January 2021)

• Record Type: Journal Article
• Authors: Suh, Sungho; Lee, Haebom; Lukowicz, Paul; Lee, Yong Oh
• Is Part Of: Neural networks. Volume 133(2021)
• Page Start: 69
• Page End: 86
• Publication Date: 2021-01
• Subjects: Imbalanced classification -- Data augmentation -- Generative adversarial networks -- Classification enhancement -- Ambiguous classes; Neural computers -- Periodicals; Neural networks (Computer science) -- Periodicals; Neural networks (Neurobiology) -- Periodicals; Nervous System -- Periodicals; Ordinateurs neuronaux -- Périodiques; Réseaux neuronaux (Informatique) -- Périodiques; Réseaux neuronaux (Neurobiologie) -- Périodiques; Neural computers; Neural networks (Computer science); Neural networks (Neurobiology); Periodicals; 006.32
• DOI: 10.1016/j.neunet.2020.10.004
• Languages: English
• ISSNs: 0893-6080

Abstract

Abstract: The data imbalance problem in classification is a frequent but challenging task. In real-world datasets, numerous class distributions are imbalanced and the classification result under such condition reveals extreme bias in the majority data class. Recently, the potential of GAN as a data augmentation method on minority data has been studied. In this paper, we propose a classification enhancement generative adversarial networks (CEGAN) to enhance the quality of generated synthetic minority data and more importantly, to improve the prediction accuracy in data imbalanced condition. In addition, we propose an ambiguity reduction method using the generated synthetic minority data for the case of multiple similar classes that are degenerating the classification accuracy. The proposed method is demonstrated with five benchmark datasets. The results indicate that approximating the real data distribution using CEGAN improves the classification performance significantly in data imbalanced conditions compared with various standard data augmentation methods. Highlights: A classification enhancement generative adversarial networks (CEGAN) is introduced to improve the classification under the imbalanced data condition. The proposed method is composed of three independent networks, a generator, a discriminator, and a classifier. By designing a loss function for ambiguous classes, we propose a classification enhancement GAN for ambiguity reduction (CEGAN-AR). The proposed method outperforms various standard data augmentation methods under data imbalanced conditions.