Feature selection using firefly optimization for classification and regression models.

In this research, we propose a variant of the Firefly Algorithm (FA) for discriminative feature selection in classification and regression models for supporting decision making processes using data-based learning methods. The FA variant employs Simulated Annealing (SA)-enhanced local and global promising solutions, chaotic-accelerated attractiveness parameters and diversion mechanisms of weak solutions to escape from the local optimum trap and mitigate the premature convergence problem in the original FA algorithm. A total of 29 classification and 11 regression benchmark data sets have been used to evaluate the efficiency of the proposed FA model. It shows statistically significant improvements over other state-of-the-art FA variants and classical search methods for diverse feature selection problems. In short, the proposed FA variant offers an effective method to identify optimal feature subsets in classification and regression models for supporting data-based decision making processes.