Application of GAN-based Data Augmentation in Imbalanced Classification of Industrial Internet of Things

This study focuses on the imbalanced classification problem in the Industrial Internet of Things (IIoT). Due to the scarcity of fault samples, traditional classifiers often show poor performance. To address this, a data augmentation method based on Generative Adversarial Networks (GANs) is proposed. The designed GAN architecture combines LSTM and Transformer in the generator to capture long - term dependencies and global features in time - series data, which is typical in IIoT scenarios. The discriminator uses spectral normalization to enhance stability. A two - stage training strategy is adopted to ensure the generation of high - quality samples. Dynamic threshold screening and mixing proportion optimization are introduced to balance the data distribution and improve the classifier's recognition ability for minority classes. Experiments on NASA bearing fault, TE chemical process, and PHM gearbox fault datasets demonstrate that the proposed method significantly outperforms traditional oversampling and generation methods in extremely imbalanced situations. It can improve the F1 - Score by up to 30% and achieve an AUC - ROC of 0.91 - 0.93, verifying its robustness and generalization ability. This research provides a practical solution for equipment health management in the IIoT, with potential applications in predictive maintenance and digital twin systems.