Edge-Cloud Differential Co-Evolutionary Algorithm for Distributed Feature Selection Optimization

Evolutionary algorithms have become a popular method for feature selection to reduce the dimensionality of data. However, with the rapid development of distributed computing paradigms, data are owned by distributed nodes, which poses challenges for global optimization. Meanwhile, the rapid growth of data size increases the burden of a large number of fitness evaluations. To solve these problems, this paper proposes an edgecloud differential co-evolutionary algorithm for distributed highdimensional feature selection optimization. The edge clients are responsible for local model construction and local optimization, while the cloud server takes charge of global model ensemble and global optimization. Specifically, the edge clients train a radial basis function network using local data and conduct local evolution through differential evolution (DE). The training error and local optimized candidates are sent to the cloud server through the proposed best-of-the-best communication mechanism. After receiving the local information, the cloud server ensembles the local models for global optimization. Meanwhile, the local candidates are integrated into a global solution to guide the global DE evolution. Additionally, the final iteration begins to integrate the edge client solution set and find a federated set that feeds back to the cloud server to achieve the 'best of the best' effect. Experimental results demonstrate that the proposed algorithm is superior to traditional feature selection and integration methods on five data sets.