The quick expansion of IoT devices has produced an incredible amount of data, and sophisticated machine learning solutions are frequently needed to extract valuable insights. However, IoT data analytics is severely hindered by privacy issues, unbalanced datasets, and the scarcity of labeled data. To overcome these constraints, this paper investigates using Generative Adversarial Networks (GANs) to produce high-quality synthetic IoT data. To address the issues of mode collapse and unstable training that plague conventional GANs, this paper develops a sophisticated enhanced GAN architecture that includes self-attention mechanisms, gradient penalty stabilization, and Wasserstein loss. The proposed EGAN-IoT was trained on the IoT-23 dataset, a widely used benchmark for IoT traffic analysis, and evaluated using quantitative metrics such as Fréchet Inception Distance (FID) and Wasserstein Distance, along with visual correlation analyses. The results demonstrate significant improvements in the fidelity and diversity of generated data compared to Vanilla GAN, DCGAN, and WGAN-GP architectures. The generated synthetic data preserves key statistical properties of the real data, validating its potential for downstream applications like anomaly detection and data augmentation. This research provides a robust framework for synthetic IoT data generation, bridging the gap between data scarcity and the growing demands of IoT analytics.