Enhanced Tree-Seed Algorithm with Double-Layer Cooperation Strategy to Boost Diversity and Exploration Capability for Feature Selection

Tree-Seed optimization algorithm (TSA) has certain advantages in solving optimization problems, and the seed generation mechanism can ensure that the global optimal solution is recovered. However, the update method for the tree is a single process that relies solely on the generation and replacement of the seed to complete the update for the parent tree. This approach will inevitably cause the problems of the imbalance between exploration and exploitation, poor population diversity and local stagnation. Inspired by the Butterfly Optimization Algorithm (BOA), a novel hybrid Tree-Seed via Butterfly Optimization Algorithm (TSBA) with double-layer cooperation strategy, including seed-layer with triple-based self-adaption mechanism and Tree-layer with fragrance-guided mechanism, is proposed to improve exploration capability. The proposed TSBA algorithm is then tested on IEEE CEC 2017 benchmark functions, comparing with three types of algorithms, followed by the Wilcoxon signed-rank and Friedman ranking with post-hoc tests. Moreover, the applicability for feature selection is verified by designing bTSBA as a binary version 10 UCI datasets. All experimental results have consistently confirmed that the proposed TSBA algorithm is a promising optimization and provide an effective solution for the optimization problems.