Evaluating components of artificial immune algorithms: A Performance-aware Method based on Evolutionary Calibrator

We are interested in methods and strategies that allow us to simplify bio-inspired algorithms without reducing their accuracy. These algorithms are usually designed and implemented adding new components incrementally which makes inherently difficult to understand the relation between them and their individual contribution to the algorithm performance. In this paper, the information obtained when using a tuner to identify a set of good parameter values is analyzed and a method to use this tuner in order to help us to take design decisions is proposed. Our results are shown and our approach is validated using an artificial immune algorithm which has been proposed to solve multi-objective problems. The results show that these decisions lead to a code that is shorter than that of the initial algorithm while maintaining its performance.