Applied Soft Computing 1 (2002) 285–299 Mutation-based genetic algorithm: performance evaluation I. De Falco a,∗ , A. Della Cioppa b , E. Tarantino a a ISPAIM, National Research Council of Italy, Via Patacca 85, I-80056 Ercolano (NA), Italy b Department of Computer Science and Electrical Engineering, University of Salerno, Via Ponte Don Melillo 1, I-84084 Fisciano (SA), Italy Received 26 October 2001; received in revised form 23 January 2002; accepted 26 February 2002 Abstract The role of mutation has been frequently underestimated in the field of Evolutionary Computation. Moreover only little work has been done by researchers on mutations other than the classical point mutation. In fact, current versions of Genetic Algorithms (GAs) make use of this kind of mutation only, in spite of the existence in nature of many different forms of mutations. In this paper, we try to address these issues starting from the definition of two nature-based mutations, i.e. the frame-shift and the translocation. These mutation operators are applied to the solution of several test functions without making use of crossover. A comparison with the results achieved by classical crossover-based GAs, both sequential and parallel, shows the effectiveness of such operators. © 2002 Elsevier Science B.V. All rights reserved. Keywords: Genetic algorithms; Mutation; Frame-shift; Translocation 1. Introduction Since the 1960s much research on Evolutionary Al- gorithms (EAs) has been devoted to investigation of importance of the involved operators. In Genetic Algo- rithms (GAs) [14,18] special attention has been ded- icated to crossover, while mutation has always been seen as a secondary operator, though useful in intro- ducing diversity in the population. In Evolution Strate- gies (ESs) [30,32] and in Evolutionary Programming (EP) [11], instead, mutation has been considered as the main operator driving evolution, and special care has been taken in designing suitable models of such an operator. Another difference between those schools is in the mutation frequency. In GAs mutation is to ∗ Corresponding author. Tel.: +39-81-560-8330; fax: +39-81-613-9219. E-mail addresses: i.defalco@ispaim.na.cnr.it (I. De Falco), adellacioppa@unisa.it (A. Della Cioppa), e.tarantino@ispaim.na.cnr.it (E. Tarantino). be applied with a low probability, while in ESs and in EP it is always applied to each problem variable. Much research has been carried out in the field of GAs about the optimal mutation rate. This has led to lots of papers aiming at assessing optimal mutation rates. Independently of the approach, however, the mutation operators taken into account often seem to represent either an oversimplified model of biological mutation or a completely new one, “artificial”, specifically tai- lored to deal with real variables, without any corre- spondence with nature. During last years this problem has been evidenced by several authors. This search for more realistic mutation operators has involved many researchers, like for instance Mitchell and Forrest [23] and Banzhaf et al. [2], who have both pointed out the importance of considering new evolution operators. As a result, several operators relying on natural mech- anisms and some mutation-based models of evolution have been developed. In [5–7] we introduced two brand-new mutation operators suitable for GAs, i.e. the frame-shift and the 1568-4946/02/$ – see front matter © 2002 Elsevier Science B.V. All rights reserved. PII:S1568-4946(02)00021-2