Modified-generative adversarial networks for imbalance text classification

Textual data faces a significant challenge known as class imbalance. This challenge arises when the distribution of classes within a dataset is highly disproportionate. To address this problem, effective algorithms that can handle the imbalance in the data while maintaining high predictive accuracy are needed. To solve the class imbalance problem, numerous methods have been proposed, which include SMOTE (Synthetic Minority Over Sampling), ADASYN (Adaptive Synthetic Sampling), ensemble methods, etc. In this paper we propose the novel Synthetic Minority Over Sampling Technique Nominal Continuous-Generative Adversarial Network (SMOTE NC-GAN) Model to tackle the problem of imbalanced text datasets. This method provides a novel solution for addressing the class imbalance problem. It involves generating synthetic data from the minority class of dataset using SMOTE-NC and subsequently applying that data along with real data fed to the Generative Adversarial Network. By using a combination of SMOTE-NC and GAN techniques, our proposed method can generate synthetic samples that enhance the quality of minority classes in the text datasets. Our proposed model addresses the limitations of both GAN and SMOTE. Empirical findings on various benchmark datasets have demonstrated the superior performance of the SMOTE NC-GAN Model over different state-of-the-art methods. This model enhances sample quality and offers computational efficiency, making it a promising solution for handling imbalanced datasets across various domains. The classification performance is significantly improved when utilizing the SMOTE NC-GAN Model in comparison to the Synthetic Minority Oversampling Technique.