A PSO-based Surrogate-Assisted Federated Feature Selection Algorithm

High-dimensional feature selection (FS) problems face challenges such as the “curse of dimensionality” and high computational costs. In addition, a large amount of high-dimensional data describing the same learning task may be horizontally distributed among different institutions (called participants) and cannot be shared because of privacy protection limitations, further increasing the difficulty of high-dimensional FS problems. Therefore, a surrogate-assisted federated evolutionary FS algorithm based on particle swarm optimization (PSO) is proposed to solve the FS problem of high-dimensional data with multi-party participation under privacy protection. First, a joint filtered FS method based on the XGBoost model is proposed, which significantly reduces the initial space of features while ensuring privacy. Subsequently, a surrogate-assisted federated evolutionary FS algorithm framework under privacy protection is designed, and based on this framework, a joint construction and management strategy for the surrogate model, a joint evaluation strategy based on the surrogate model, and a joint update strategy for particles are developed. Finally, the proposed algorithm is applied to 10 test datasets, and compared with two typical evolutionary FS algorithms. The experimental results show that the proposed algorithm can not only ensure the classification performance but also significantly improve the efficiency of the algorithm while protecting the privacy of the participant data.