A novel enhanced neural network for anomaly detection in the IoT environment

Abstract

The Internet of Things (IoT) is changing many fields, including healthcare, transportation, and smart cities. However, the rise of serious cybersecurity issues. This paper introduces a new architecture of Fully connected neural networks (FCNNs) designed for detecting anomalies in IoT settings. The proposed model uses residual connections, multi-head attention mechanisms, SHAP-based feature selection, and RAdam optimisation to improve detection accuracy, robustness, and interpretability. Extensive tests on two benchmark datasets, ToN-IoT and UNSW-NB15, show that the model works well. Specifically, it achieved an accuracy of 99. 98% and an F1 score of 99. 95% in the ToN-IoT dataset, as well as an accuracy and F1 score of 99. 74% in the UNSW-NB15 dataset. These findings suggest that the improved FCNNs architecture can effectively identify anomalies while being lightweight enough for use in real-time IoT systems. The addition of explainability techniques further enhances its usefulness in critical cybersecurity situations by providing precise and interpretable predictions.