Effect of epistasis on the performance of genetic algorithms
Sajad Jafari, Tomasz Kapitaniak, Karthikeyan Rajagopal, Viet–Thanh Pham, Fawaz E. Alsaadi · Journal of Zhejiang University. Science A · 2018
In the field of genetics, it is well known that a specific genetic behavior may be influenced by more than one gene. There is a similar concept in genetic algorithms (GAs), called epistasis, which is the interaction between genes. This study demonstrates that, in spite of what is generally assumed, GAs are not an efficient optimization tool. This is because the main operator, mating (crossover), cannot function properly in epistatic optimization problems. In non-epistatic problems, although a GA can possibly provide a correct solution, it is an inefficient and time-consuming algorithm. As proof, we used conventional test functions and introduced new ones and confirmed our claim with simulation results.