Abstract

This dissertation compares the performance of five existing Genetic Algorithms (GAs) that do not require the manual tuning of their parameters, and are thus called Parameterless Genetic Algorithms (pGAs). The five pGAs selected for evaluation span the three most important categories of Parameterless GAs: Deterministic, Adaptive and Self-Adaptive pGAs. The five test functions used to evaluate the performance of the pGAs include unimodal, multimodal and deceptive functions. We assess performance in terms of fitness, diversity, reliability, speed and memory load. Surprisingly , the simplest Parameterless GA tested proves to be the best overall performer. Last, but not least, we describe a new parameterless Genetic Algorithm (nGA), one that is easy to understand and implement, and which bests all five tested pGAs in terms of performance, particularly on hard and deceptive surfaces.