Co-evolving fuzzy decision trees and scenarios

A co-evolutionary data mining algorithm has been invented that automatically generates decision logic in the form of fuzzy decision trees (FDTs). The algorithm initially uses a genetic program (GP) to mine a database of scenarios to automatically create the fuzzy logic. This is followed by the application of a genetic algorithm (GA) that is used to search for pathological scenarios (PS) that result in unsatisfactory performance by the fuzzy logic found by the GP. The fuzzy logic found in the previous step by the GP along with failure criteria (FC) is used to form the fitness function for the GA. If the GA fails to find pathological scenarios then the co-evolution ends; otherwise, the new scenarios are appended to the GPpsilas database followed by GP based data mining and a GA scenario search. A detailed description of the co-evolution of a fuzzy decision tree for real-time control of unmanned air vehicles is provided. The fitness functions for the GP, terminal set, function set, and methods of accelerating convergence are included. The fitness function for the GA and a method of representing scenarios as chromosomes are given. Simulations related to validation of the fuzzy logic are discussed.