Self-Similarity Feature Extraction Using the Hurst Parameter in Transformer Architectures for IoT Anomaly Identification

The increasing deployment of Internet of Things (IoT) devices has amplified the importance of effective anomaly detection to ensure network security and reliable operation. Traditional methods often struggle to accurately identify complex and subtle abnormal patterns within high-volume, multidimensional traffic data. In this study, we introduce a novel framework that combines Transformer-based deep learning models with self-similarity features derived from the Hurst exponent to enhance anomaly detection performance. By extracting fractal-based features and employing attention mechanisms within the Transformer architecture, our approach effectively captures long-range dependencies and dynamic traffic behaviors. Experimental results on a benchmark IoT dataset demonstrate a high detection accuracy of 95.64%, outperforming many existing methods. The proposed model not only improves detection effectiveness but also provides interpretability and robustness against noisy data. Visualizations of confusion matrices and anomaly frequency analyses further validate the effectiveness of our approach. This work highlights the potential of integrating fractal features with advanced neural architectures for real-time, reliable IoT network security.