Saved in:
| Main Author: | |
|---|---|
| Format: | Recurso digital |
| Language: | |
| Published: |
Zenodo
2026
|
| Online Access: | https://doi.org/10.5281/zenodo.20109457 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Table of Contents:
- <p>The explosive growth of edge-cloud computing, Internet of Things (IoT) infrastructures, and distributed artificial intelligence has created significant challenges related to data privacy, explainability, communication overhead, and energy sustainability. Traditional centralized machine-learning approaches often require large-scale transmission of sensitive data to cloud servers, leading to increased privacy risks, higher latency, excessive energy consumption, and dependence on non-renewable grid power. To address these limitations, this research proposes FL-XAI-ER, a novel federated learning framework that jointly integrates differential privacy, explainable artificial intelligence (XAI), renewable-energy-aware scheduling, and reinforcement-learning-driven optimization for sustainable edge-cloud intelligence. The proposed framework enables distributed edge devices to collaboratively train machine-learning models without sharing raw data, thereby preserving user privacy while maintaining high computational performance. Differential privacy mechanisms are incorporated to protect local model updates, whereas explainability modules based on LIME and SHAP generate interpretable and privacy-preserving decision explanations suitable for critical applications such as healthcare, smart grids, industrial automation, and intelligent transportation systems. Furthermore, a reinforcement-learning-based energy scheduler dynamically coordinates federated training rounds according to renewable-energy availability, battery state, communication latency, and system workload conditions. To improve scalability and communication efficiency, the framework incorporates secure aggregation, sparse gradient compression, and quantization mechanisms that significantly reduce transmission overhead while maintaining model convergence stability. Experimental evaluation using distributed edge-cloud environments and renewable-energy datasets demonstrates substantial performance improvements, including approximately 62% reduction in energy consumption, 2.1× faster convergence compared with conventional FedAvg, 133× communication compression, and nearly 68% reduction in carbon emissions. The results confirm that FL-XAI-ER provides a scalable, interpretable, privacy-preserving, and environmentally sustainable federated-learning solution for next-generation intelligent edge-cloud ecosystems.</p> <p><strong><br><br>Index Terms: Federated Learning, Explainable Artificial Intelligence (XAI), Differential Privacy, Edge-Cloud Computing, Renewable Energy Optimization, Reinforcement Learning Scheduling, Secure Aggregation.</strong></p>