Front-rear crossover: A new crossover technique for solving a trap problem

Abstract

Crossover methods are important keys to the success of genetic algorithms. However, traditional crossover methods fail to solve a trap problem, which is a difficult benchmark problem designed to deceive genetic algorithms to favor all-zero bits, while the actual solution is all-one bits. The Bayesian optimization algorithm (BOA) is the most famous algorithm that can solve the trap problem; however, it incurs a large computational cost. This paper, therefore, proposes a novel crossover technique, called a front-rear crossover (FRC), to enhance the simple genetic algorithm. We test the proposed technique with various benchmark problems and compare the results with four other crossover algorithms, including single point crossover (SPC), two point crossover (TPC), uniform crossover (UC) and ring crossover (RC). The FRC outperforms the four techniques in all test problems. It can also solve the trap problem by requiring the 40 times lesser number of fitness evaluations than BOA's.