An Alternative Archiving Technique for Evolutionary Polygonal Approximation

Abstract

Archiving procedures are a key parameter for Multi-objective evolutionary algorithms, since they guarantee the algorithm convergence and the good spread of the obtained solutions in the final Pareto front. For many practical applications, the cost of the algorithm is clearly dominated by the computational cost of the underlying fitness functions, allowing complex processes to be incorporated into the archiving procedure. This work presents a study of the archiving technique for evolutionary polygonal approximation (the division of a given curve into a set of n segments represented by a linear model) based on the epsilon-glitch concept, highlighting the cost of the technique compared to the fitness computation, and proposing a novel alternative archiving procedure, which yields statistically significant better results compared to available approaches.