XAI-EdgeSFL: Explainable Edge Intelligence With Adaptive Intrusion-Resilient Split Federated Learning for Consumer Healthcare Ecosystems

The fast growth of consumer electronics, particularly Internet of Things (IoT) devices and wearables, has significantly changed healthcare systems by enabling for real-time monitoring and customized medical treatment. Edge computing is changing healthcare by enabling the secure and distributed processing of data on consumer devices. However, this integration presents substantial security issues, raising significant concerns regarding AI-driven decision-making, model understandability, and privacy protection. This paper introduces XAI- EdgeSFL, a novel framework that integrates Explainable Artificial Intelligence (XAI) with an Adaptive Intrusion-Resilient Split Federated Learning (SFL) architecture to provide reliable and transparent edge health monitoring. The SFL loop computes important features of sensitive health data using a lightweight Temporal Convolutional Network (TCN) on edge devices and a more complex classifier on a global server. This method enhances privacy by retaining raw data on the device. In addition, we present an adaptive aggregation approach that dynamically weights client model updates to evaluate intrinsic resilience against emerging cyberthreats. SHapley Additive exPlanations (SHAP), Local Interpretable Model-agnostic Explanations (LIME), and Integrated Gradient (IG) are a multi-modal XAI suite that offers feature-level explanations for clinical trust. The proposed method performs well across all datasets, achieving 86.15% accuracy and 86.01% F1-score on CICIoMT2024, 93.73% accuracy and 93.65% F1-score on N-BaIoT, and 99.46% accuracy and 99.47% F1-score on CICIDS2017. The results show that XAI-EdgeSFL improves security, privacy, and transparency for AI-driven consumer healthcare systems.