A Pre-initialization Stage of Population-Based Bio-inspired Metaheuristics for Handling Expensive Optimization Problems

Abstract

Metaheuristics are probabilistic optimization algorithms which are applicable to a wide range of optimization problems. Bio-inspired, also called nature-inspired, optimization algorithms are the most widely-known metaheuristics. The general scheme of bio-inspired algorithms consists in an initial stage of randomly generated solutions which evolve through search operations, for several generations, towards an optimal value of the fitness function of the optimization problem at hand. Such a scenario requires repeated evaluation of the fitness function. While in some applications each evaluation will not take more than a fraction of a second, in others, mainly those encountered in data mining, each evaluation may take up several minutes, hours, or even more. This category of optimization problems is called expensive optimization. Such cases require a certain modification of the above scheme. In this paper we present a new method for handling expensive optimization problems. This method can be applied with different population-based bio-inspired optimization algorithms. Although the proposed method is independent of the application to which it is applied, we experiment it on a data mining task.