The application of genetic algorithms in behavioural ecology, illustrated with a model of anti-predator vigilance.

We develop a genetic algorithm (GA) approach to a well-known model of vigilance behaviour in a group of animals. We first demonstrate that the GA approach can provide a good match to analytic solutions to the original model. We demonstrate that a GA can be used to find the evolutionarily stable strategies in a model relevant to behavioural ecology where the fitness of each strategy is determined by the frequencies of different strategies in the population. We argue that the GA implementation demonstrates the combination of assumptions used to generate analytic solution to the original model can only be simultaneously satisfied under relatively restrictive conditions on the ecology of the species involved; specifically that group membership is very fluid but group size is conserved over timescales of individual foraging bouts. We further explore the sensitivity of model predictions to alternative choices in the implementation of the GA, and present advice for implementation and presentation of similar models. In particular, we emphasise the need for care in measuring the predictions of such models, so as to capture the intrinsic behaviour of the system and not the remnant of often arbitrarily chosen initial conditions. We also emphasise the potential for GA models to be more transparent about model assumptions regarding underlying biology than analytic models.