Evaluation of Reverse Ellitism Firefly Algorithm on the CEC2013 Real Parameter Single Objective Optimization Benchmark Functions

This paper proposes a modified firefly algorithm (RE-FA), namely Reverse Ellitism Firefly Algorithm. The primary objective of this algorithm is faster convergence around global minima for training of Neural Networks in continious domain. The secondary objective of this modified Firefly Algorithm was to consider all parameters namely α, β,γ to be adaptive in accordance to the objective function. No a-priori knowledge of the problem domain is required, as such; no preset or tuning of the said parameters is required beforehand. A failsafe criterion for clearance of local minima is also applied. The tendency of standard Firefly algorithm to get stuck in local minima and slow convergence is mitigated by applying reverse ellitism in RE-FA.