Federated learning enabled multi-key homomorphic encryption
The rapid advancement of digital healthcare systems has significantly increased the volume of medical data, raising concerns about its security and confidentiality. As patient information is highly sensitive, any compromise could have severe consequences in healthcare delivery. Federated Learning (F...
Uloženo v:
| Vydáno v: | Expert systems with applications Ročník 268; s. 126197 |
|---|---|
| Hlavní autoři: | , |
| Médium: | Journal Article |
| Jazyk: | angličtina |
| Vydáno: |
Elsevier Ltd
05.04.2025
|
| Témata: | |
| ISSN: | 0957-4174 |
| On-line přístup: | Získat plný text |
| Tagy: |
Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
|
| Shrnutí: | The rapid advancement of digital healthcare systems has significantly increased the volume of medical data, raising concerns about its security and confidentiality. As patient information is highly sensitive, any compromise could have severe consequences in healthcare delivery. Federated Learning (FL) has emerged as a promising solution by enabling decentralized data processing through local training and global model aggregation. Despite its advantages, FL faces challenges in ensuring the secure transmission of data and preserving privacy during model updates. Homomorphic encryption (HE)-based security is a promising approach for secure data transmission. Hence, the Puzzle Archimedes Optimization Algorithm (PAOA)-based multi-key HE is designed in this research. The healthcare data is obtained from the dataset and shared among many parties using a multi-party data-sharing platform. All parties utilize noise-free HE to protect sensitive data and enhance communication security. The ideal multi-keys are created using the proposed PAOA. Moreover, the PAOA-trained deep residual network (DRN) (PAOA_DRN) is utilized for privacy data classification. The outcome from each of the local training is aggregated into the global training, and the weights of every local training model are updated. Furthermore, the training accuracy, training loss, mean average precision, and Mean Squared Error (MSE) metrics are utilized for computing the efficacy of the PAOA_DRN. On the Cleveland dataset, the model achieved superior results with a training accuracy of 0.946, training loss of 0.053, MAP of 0.951, and MSE of 0.182. This research provides a significant step forward in addressing critical issues in healthcare data security, with the potential for broad application in medical data sharing and privacy-preserving analytics. |
|---|---|
| ISSN: | 0957-4174 |
| DOI: | 10.1016/j.eswa.2024.126197 |