FedEdge AI-TC: A Semi-supervised Traffic Classification Method based on Trusted Federated Deep Learning for Mobile Edge Computing

Privacy protection through local processing: The framework uses federated learning to train AI models locally on individual devices rather than sending raw network data to central servers, addressing a key privacy concern in smart home security systems.

Reduced need for labeled data: By combining semi-supervised learning techniques (VAE and CNN), the system can learn effectively with fewer manually labeled examples, lowering the resource burden and cost of preparing training data for network traffic classification.

Explainability improves trustworthiness: The paper introduces "XAI-Pruning," a method that makes AI decision-making more transparent and interpretable, helping address the "black box" problem that can undermine user trust in automated security systems.

Efficient operation on resource-limited devices: The model compression techniques enable sophisticated AI-based traffic classification to run directly on 5G home gateways and similar edge devices with limited computing power, rather than requiring constant communication with distant data centers.

Balancing multiple competing demands: The framework simultaneously tackles four major challenges—privacy concerns, data scarcity, computational limitations, and model transparency—demonstrating that responsible AI implementation requires addressing security, efficiency, and explainability together rather than in isolation.