A machine learning framework combining Conditional Tabular Generative Adversarial Networks with One-Dimensional Convolutional Neural Networks improves minority class classification accuracy across imbalanced datasets without compromising majority class performance, researchers report in Neural Computing and Applications. Across extensive benchmark evaluations, the proposed method outperformed state-of-the-art classification algorithms by generating higher Recall, F1-Score, and G-Mean values for the minority class, thereby establishing balanced classification performance across both data classes.
The investigation addresses a persistent challenge in practical classification problems such as financial fraud detection, medical diagnosis, and intrusion detection, where data collections are typically highly imbalanced. In these specialized operational domains, the data class containing the least number of instances often carries the most critical information. Conventional classification algorithms, however, are effective for classifying the majority class but perform poorly when handling the minority class.
Standard historical approaches to addressing this severe imbalance typically involve either oversampling the minority class or undersampling the majority class in an effort to improve minority classification accuracy. Nevertheless, the authors emphasize that these conventional data sampling strategies frequently introduce unwanted noise or induce algorithmic overfitting, difficulties that emerge as especially severe when working in complex, high-dimensional spaces.
To address these documented shortcomings, the research team proposed an innovative methodology integrating Conditional Tabular Generative Adversarial Networks, designated as CTGANs, with One-Dimensional Convolutional Neural Networks, or 1D-CNNs. Under this collaborative pipeline, the CTGAN architecture is implemented specifically to conduct targeted data augmentation for the minority class, while the 1D-CNN structure is employed directly for the final classification task.
The researchers validated their approach by carrying out extensive comparative experiments across 50 benchmark datasets sourced from the University of California, Irvine (UCI) Machine Learning Repository. The resulting experimental outcomes demonstrated that the proposed method consistently surpassed state-of-the-art classification algorithms, achieving higher values of Recall, F1-Score, and G-Mean for the minority class while preserving majority class accuracy without compromise.
Subject of Research: Minority class data augmentation and classification using CTGANs and 1D-CNNs
Article Title: Improving minority class classification accuracy using conditional tabular generative adversarial networks
Article References: Dutta, D., Mondal, S., Mukherjee, D., & Goswami, K. (2026). Improving minority class classification accuracy using conditional tabular generative adversarial networks. Neural Computing and Applications, 38(22), Article 895. https://doi.org/10.1007/s00521-026-12567-1
Image Credits: AI Generated
DOI: 10.1007/s00521-026-12567-1
Keywords: conditional tabular generative adversarial networks, CTGAN, one-dimensional convolutional neural networks, 1D-CNN, minority class classification, imbalanced data, data augmentation, UCI Machine Learning Repository, Recall, F1-Score, G-Mean, financial fraud detection, medical diagnosis, intrusion detection, oversampling, undersampling, high-dimensional spaces
Cite Scienmag News
Denise Maddox. (November 16, 2026). Generative Adversarial Networks and Convolutional Models Improve Imbalanced Data Classification Accuracy. Scienmag. https://scienmag.com/generative-adversarial-networks-and-convolutional-models-improve-imbalanced-data-classification-accuracy/
Denise Maddox. "Generative Adversarial Networks and Convolutional Models Improve Imbalanced Data Classification Accuracy." Scienmag, 16 November 2026, https://scienmag.com/generative-adversarial-networks-and-convolutional-models-improve-imbalanced-data-classification-accuracy/. Accessed 16 November 2026.
Denise Maddox. "Generative Adversarial Networks and Convolutional Models Improve Imbalanced Data Classification Accuracy." Scienmag. November 16, 2026. https://scienmag.com/generative-adversarial-networks-and-convolutional-models-improve-imbalanced-data-classification-accuracy/








