XAI-SOH-FL: Enhancing SOH-FL with Adaptive Aggregation and Explainable AI for Intrusion Detection in Heterogeneous IoT

Researchers propose XAI-SOH-FL, an enhanced federated learning framework for IoT intrusion detection that combines adaptive aggregation mechanisms with explainable AI to address data heterogeneity and model interpretability challenges. The system achieves 94.12% accuracy on benchmark datasets while eliminating manual parameter tuning and providing transparent feature-level insights into security decisions.

XAI-SOH-FL represents a meaningful advancement in securing heterogeneous IoT networks by addressing three critical pain points in federated learning-based intrusion detection. The framework's integration of adaptive aggregation through similarity thresholding and Bayesian Optimization automates parameter selection, reducing operational overhead while improving model robustness across distributed environments with varying data distributions.

The inclusion of SHAP-based explainability addresses a significant gap in IoT security infrastructure, where stakeholders increasingly demand transparency in automated threat detection. Traditional black-box models create liability and compliance risks, particularly in regulated sectors. By identifying that flow-level features like packet length and flow duration drive detection decisions, the framework enables security teams to understand and validate model behavior, building institutional trust in federated systems.

For IoT ecosystem developers and enterprise security teams, this work demonstrates that privacy-preserving approaches no longer require sacrificing interpretability or accuracy. The 94.12% accuracy and 0.92 F1-score metrics position federated learning as a viable alternative to centralized approaches, particularly for organizations handling sensitive network data across multiple facilities or jurisdictions. The reduced communication rounds indicate improved computational efficiency, directly lowering bandwidth costs in resource-constrained IoT deployments.

The practical impact extends to supply chain security and critical infrastructure protection, where heterogeneous device networks demand adaptive threat detection. As regulatory frameworks around AI transparency tighten globally, frameworks like XAI-SOH-FL address compliance requirements while maintaining distributed privacy guarantees. Future research should explore deployment on actual edge hardware and evaluation against zero-day attack patterns.

• →XAI-SOH-FL achieves 94.12% accuracy and 0.92 F1-score while eliminating manual aggregation parameter tuning in federated IoT intrusion detection.
• →Adaptive gamma selection mechanism automatically optimizes model aggregation across heterogeneous data distributions without manual intervention.
• →SHAP-based explainability reveals flow-level features like packet length and flow duration as primary drivers of intrusion detection decisions.
• →Framework converges in fewer communication rounds than baseline SOH-FL, reducing bandwidth overhead for distributed IoT deployments.
• →Approach balances privacy preservation, model accuracy, and interpretability—addressing compliance and trust requirements in enterprise IoT security.