An Ensemble Framework to Forest Optimization Based Reduct Searching

Essentially, the solution to an attribute reduction problem can be viewed as a reduct searching process. Currently, among various strategies, meta-heuristic has received extensive attention. As new emerging approach, forest optimization algorithm (FOA) is introduced solving of in this study. To further improve classification performance selected attributes reduct, ensemble framework also developed: firstly, multiple reducts are obtained by FOA and data perturbation, structure those symmetrical, which indicates that no order exists reducts; secondly, used execute voting over testing samples. Finally, comprehensive experiments on 20 UCI datasets clearly validated effectiveness our framework: it not only beneficial output with superior accuracies stabilities but suitable for pre-processing noise. This improvement work we have performed makes obtain better benefits processing life, health, medical other fields.