TrustFed-RHIO: an optimization-driven differential privacy federated learning framework for secure and explainable...

A rapid proliferation of industrial internet of things (IIoT) systems has increased the vulnerability of interconnected devices for sophisticated cyberattacks, which necessitates intelligent and privacy-preserving solution for security. This paper presents TrustFed-RHIO, a novel hybrid model which integrates rock hyrax intelligence optimization (RHIO) for optimal selection of feature with a trustworthy differential privacy-enhanced federated learning (TrustFed) scheme for the collaborative detection of attack. The algorithm of RHIO mimics collective intelligence of rock hyrax colonies for identifying most discriminative features, thus reducing dimensionality and enhancing efficiency of classifier. The proposed TrustFed-RHIO scheme ensures, secure, distributed learning over multiple IIoT nodes on embedding differential privacy mechanisms thus mitigating the data leakage risks and adversarial inference. A suggested scheme is thus empowered with explainable artificial intelligence (XAI) scheme termed SHAP and LIME for enhancing interpretability and trust in model predictions. Experimental estimation on benchmark IIoT dataset shows that TrustFed-RHIO attains superior performance on detection accuracy, robustness against adversarial attacks, and privacy preservation on comparing existing schemes. At last, this framework supports secure storage of cloud on detection outcomes, thus enabling scalable deployment in the real-world IIoT framework. The performance evaluation is carried on benchmark dataset CCIoT2024-DIAD and performance is estimated for various metrics like latency, accuracy, precision, recall, F1-score, specificity, training time, and so on. Overall analysis shows that the proposed model is effective in detecting IIoT attacks.