A BSO-based Feature Selection Algorithm Under Privacy Protection
Ying Hu, Wanqiu Zhang, Yong Zhang · 2022
In many practical problems, data describing the same task may be stored among different participants in decentralized form. Considering the case that different features are stored in different participants and data held by each participant can not be shared with each other, a vertical federated feature selection algorithm with brain storm optimization (BSOFFS) under privacy protection is proposed. BSOFFS combines the feature selection process into the SecureBoost framework, whose purpose is to reduce the dimension of distributed data jointly without sharing the local original data between participants, so that the obtained feature subset has the same or better performance with the original feature set. Finally, experimental results on several test problems show that BSOFFS can improve the classification performance of the selected feature subset remarkably while protecting data privacy.