Multi-group overlapping weighted random forests

This paper introduces an enhanced random forest framework that partitions trees into overlapping subsets, allowing each tree to contribute to multiple groups. Each group acts as a base classifier, producing predictions through internal voting, while a weighted inter-group vote combines these outputs according to each group’s reliability. A particle swarm optimization algorithm jointly determines the optimal number of groups, their composition, and associated weights, enabling efficient exploration of the configuration space. Experiments on twenty five UCI benchmark datasets show that the proposed method consistently improves accuracy, robustness, and generalization.