Securing Smart IoT Networks from Cyber Threats Using Explainable Zero Channel Attention-aided Ghost Convolution Neural Network Framework

Abstract

The rise of advancements driven by the Internet of Things (IoT), presents numerous benefits, such as improved efficiency and enhanced quality of life. However, this interconnectedness also exposes critical vulnerabilities, making robust cyber-attack detection essential. Hence, this manuscript emphasizes the innovative explainable deep learning (XAI-DL) model for detecting and classifying multiclass cyber threat attacks in Internet of Things (IoT) platforms. Initially, the raw data samples collected from the BCCC-CIC-IDS dataset are preprocessed by performing a Pareto Scaling Normalization (PSN) and one-hot encoding processto improve the data quality.After preprocessing, the Zero Channel Attention-aided Ghost Convolution Neural Network (ZCAtt-GCNN) is proposed to detect and classify the various cyber threat attacks like Denial of Service (DoS) Attacks, Distributed Denial of Service (DDoS) Attacks, Web Attacks, file transfer protocol (FTP) Attacks, and Botnet Attacks. Furthermore, three XAI models are investigated for enhanced visualizations over the cyberattack detection: Shapley additive explanations (SHAP), Partial Dependence Plot-Individual Conditional Expectation (PDP-ICE), and Permutation Feature Importance (PFI). The proposed method is simulated via the Python platform and various performance measures like G-mean, Accuracy, Matthews Correlation Coefficient (MCC), Negative Predictive Value (NPV), computation time (CT), and false positive rate (FPR) are scrutinized, and associated with other techniques. The overall accuracy of 99.48%, G-mean of 99.18%, and FPR of 0.322 are obtained by the suggested framework for identifying various IoT Cyberattacks.