Investigating Structural Bias in Real-Coded Genetic Algorithms
Kanchan Rajwar, Yogesh Kumar, Kusum Deep · Proceedings of the Genetic and Evolutionary Computation Conference Companion · 2024
Real-Coded Genetic Algorithms (RCGAs) are a significant area of focus in evolutionary algorithm research. Structural bias (SB), a recently recognized attribute in metaheuristic algorithms, leads populations to repeatedly exploit specific region(s) of the search space without acquiring new information. This behavior not only escalates the computational cost but also slows the convergence rate. While recent studies have revealed an inclination toward the center of the search space in basic Genetic Algorithms (GA), similar investigations for RCGA variants remain unexplored. This study addresses this gap by examining the bias in popular and widely used crossovers and mutations in RCGAs. The BIAS toolbox, a recently developed tool, is used for a preliminary assessment of biases in these variants. Furthermore, GST, a simple yet effective methodology, is utilized for a more thorough analysis, quantifying, and scrutinizing biases in RCGA variations. The outcomes of this research aim to deepen our theoretical understanding of RCGAs, thus enhancing their practical applications and contributing to the future development of more equitable and unbiased algorithms.