Self adaptive cuckoo search: Analysis and experimentation

Cuckoo search (CS) algorithm, based on the brood parasitic behaviour of cuckoo species, since its inception, has proved its worth in various fields of research and can be considered as an efficient problem solver. Though CS is a very good algorithm, its performance degrades as the problem complexity increases. So a new version namely self-adaptive CS (SACS) is proposed to improve its performance. The algorithm employs adaptive parameters and hence no parameter tuning is required to be done. Here proportional population reduction based on the fitness of current best and previous best solution is followed. Secondly, in order to improve the exploration and exploitation tendencies, the Gaussian sampling mechanism known as bare-bones variant along with division of population and generations is added. The concept of Weibull distributed probability switching is also added to increase the balance between the exploration and exploitation processes. Further, CS as an extension of differential evolution (DE), genetic algorithm (GA), stability analysis with respect to Von Neumann’s stability criteria are also presented. For performance evaluation, the SACS algorithm is tested on CEC 2017 benchmark problems and compared with respect to self-adaptive DE (SaDE), JADE, success-history based adaptive DE (SHADE), SHADE with linear population size reduction (LSHADE), mean-variance mapping optimization (MVMO), cuckoo version 1.0 (CV1.0), cuckoo version new (CVnew ) and others. Experimental results and statistical analysis show that the proposed SACS algorithm performs better than SaDE, JADE, CV1.0, CVnew and CS whereas comparable with respect to LSHADE, SHADE, and MVMO. Convergence profiles further validate the results.