Performance measures for niching algorithms
Jonathan Mwaura, Andries Petrus Engelbrecht, Filipe V. Nepocumeno · 2016
Numerous niching algorithms have been developed to find multiple optima for multimodal optimisation problems. Analysis of niching algorithms presents a problem due to the multiple objectives of niching algorithms, that is, to find as many optima as possible, to find accurate solutions, and to find a diverse set of solutions. Performance analysis becomes more difficult when no knowledge of the fitness landscape is available. This is due to the assumption of most performance measures that information about the optima is available. This paper provides a critical review of existing performance measures for niching algorithms. In addition, the paper introduces a new approach, the mid-point technique, to determine a unique set of solutions for performance measurement. The performance measures are used to evaluate the performance of seven particle swarm optimisation niching algorithms. The obtained results are then used to compute the predictive capability of non-knowledge assuming performance measures to that of knowledge assuming measures. The analysis shows that the considered performance measures do show a convergent behaviour.