Real options approach to evaluating genetic algorithms

Abstract

The real options technique has emerged as an evaluation tool for investment under uncertainty. It explicitly recognizes future decisions, and the exercise strategy is based on the optimal decisions in future periods. This paper employs the optimal stopping policy derived from real options approach to analyze and evaluate genetic algorithms, specifically for the new branches namely Estimation of Distribution Algorithms (EDAs). As an example, we focus on their simple class called univariate EDAs, which include the population-based incremental learning (PBIL), the univariate marginal distribution algorithm (UMDA), and the compact genetic algorithm (cGA). Although these algorithms are classified in the same class, the characteristics of their optimal stopping policy are different. These observations are useful in answering the question “which algorithm is suitable for a particular problem”. The results from the simulations indicate that the option values can be used as a quantitative measurement for comparing algorithms.