Privacy-Preserving Vertical Federated Logistic Regression for Power Data Sharing

As a typical data element, the open of the electricity data has become an inevitable trend. However, frequent data exchanges in joint analysis can lead to uncontrollable privacy and security risks, as well as significant computational and communication overhead. To address it, this paper proposes a Vertical Federated Logistic Regression algorithm based on Secret Sharing (SSVF-LR). By randomly decomposing the intermediate data exchanged during the joint analysis into multiple secret shares, the algorithm achieves low complexity of privacy computation. This approach is beneficial for promoting the secure and open circulation of electricity big data. Finally, the proposed algorithm is validated on real electricity dataset. Experimental results show that, compared to the centralized logistic regression algorithm based on the single electricity data, SSVF-LR can effectively enhance the predictive performance, while the cost time is superior to the traditional federated logistic regression algorithm based on homomorphic encryption.