Assessment of Complex Adaptive System Changeability Using a Learning Classifier System

Abstract

Although many books and conference reports have considered the changeability of a complex adaptive system (CAS), no in-depth study has assessed how changes impact a CAS or the relationships among the CAS changeability parameters of agility, flexibility, robustness, and adaptability. This study analyzes the impact of CAS changeability when an external agent requires a change and analyzes how such a change affects the evolution of the CAS. We start by reviewing the general concepts of changeability and CAS, followed by an analysis of their relationship. A model using an extension of learning classifier system (XCSF) is presented and evaluated to meet the objectives of this research to: 1) accurately assess the impact of changeability on telecommunication-based CAS components and their evolution; and 2) gain insight into the impact of changes on CAS for the purpose of improving the engineering of such systems in the future. The relationship between changeability and fitness, which is an important CAS performance measure, is included. CAS simulations using the XCSF model are compared with data from a telecommunication company over the past few years; the comparison suggests that the XCSF approach may be useful for improving CAS engineering.