Wavelet-Based Intelligent Framework for Network Traffic Anomaly Detection in IoT Embedded Systems

The rapid growth of the Internet of Things (IoT) and embedded systems has increased the vulnerability of network infrastructures to cyber-attacks, necessitating efficient real-time anomaly detection methods. In this study, we introduce a primary IoT network traffic dataset generated through controlled simulation of normal and malicious behaviors. Both time-domain features, including packet size, inter-arrival time, protocol type, and TCP flags, and frequency-domain features, such as spectral entropy and band energy derived via wavelet transform, were extracted to capture comprehensive traffic characteristics. These features were used to train a hybrid deep learning model, the Adaptive Differential Evolution Weighted Deep Belief Network (ADE-WDBN), which combines deep hierarchical feature learning with evolutionary weight optimization to enhance detection accuracy and computational efficiency. Experimental results demonstrate that ADE-WDBN outperforms traditional machine learning models and conventional deep learning approaches, achieving an accuracy of 98.37%, precision of 97.65%, recall of 98.02%, and F1-score of 97.83%. The low variability in performance across cross-validation folds indicates the model's robustness and generalizability. This research contributes a novel IoT traffic dataset and a cost-effective, adaptive anomaly detection framework capable of detecting subtle anomalies.