To be successful, a search algorithm needs to balance exploration and exploitation. In Genetic Algorithms (GAs) this is achieved through proportionate selection of individuals and reproduction operators. GAs can suffer premature convergence, when the diversity of the population decreases over time and search gets trapped in a local optimum, returning a very poor quality solution. Mask-fill reproduction operators utilizes bit-masking oriented data structure (BMODS) that maintains a good ratio between exploration and exploitation. Mask-fill reproduction operators have been utilised in various applications, but the exploration and exploitation ability of mask-fill reproduction operators have not been compared against other reproduction operators. This paper describes a rigorous and practical statistical methodology for the exploratory analysis of the mask-fill reproduction operators. First, the issues of robust experimental design and setting the control parameters for implementing a GA is addressed. Second, the impact of various reproduction operator combinations are analysed. In this study, three crossover operators and five mutation operators are considered which creates fifteen crossover-mutation operator combinations. Third, the methodology is demonstrated by considering grammatical inference (GI) problem as domain of inquiry. A hybrid genetic algorithm integrated with Sequitur algorithm (GAWS) is proposed for GI. Numerical results are presented to describe the effect of crossover-mutation operator combinations. The performance of the proposed GAWS is compared against the state-of-the-art algorithms. Statistical test are conducted to determine the performance significance of both crossover-mutation operator combinations and the proposed GAWS algorithm.
- Bit-Mask-Oriented Genetic Algorithm
- Mask-fill Crossover Operator
- Mask-fill Mutation Operator
- Genetic Algorithm
- Exploration and Exploitation.