An Evolutionary Process for Designing and Maintaining a Fuzzy Instance-based Model (FIM)

We illustrate the typical life cycle of a fuzzy knowledge-based model, from its development, testing, optimization, and deployment, through the maintenance of its knowledge base. After a brief review of related work on a static problem, the underwriting of insurance applications, we focus on the design, implementation, and updating of a Fuzzy Instance-based Model (FIM) for prediction and classification in a dynamic environment, illustrated by a fleet selection problem. After formalizing the FIM, we describe its evolutionary-based design process and evaluate its performance. Within this example of asset selection, we show the accuracy of the evolved FIM’s and their performance loss when we did not update them every six months. Finally, we advocate the use of evolutionary search intertwined with local search to further improve model life cycle.