LAUSR

The selection process is an important part of any optimization algorithm. Usually, an efficient selection process should balance between exploration of the search space and exploitation of the current knowledge about the best solutions. Cuckoo search (CS) is a simple yet powerful optimization algorithm inspired by the parasitic reproduction behavior of some cuckoo species. At each iteration of the original CS algorithm, the selection process is triggered in three places: (i) cuckoo selection where a cuckoo is selected from the population of n nests (stored solutions) based on a uniformly random function, (ii) host selection where a nest is chosen randomly from the n nests and (iii) greedy selection of a portion of the n nests for replacements with new randomly generated solutions. This paper proposes several variations of the CS algorithm by replacing the uniformly random-based selection method (used in step i) with existing randomized selection schemes, namely greedy, proportional, exponential, $$\varepsilon $$ε-greedy, softmax and reinforcement learning selection schemes. The proposed variations were evaluated and compared using twenty well-known benchmark functions (12 test functions from CEC 2005). The experimental results show that the proposed variations outperform the original CS algorithm.