Can a payoff-based learning classifier system achieve optimal performance? This paper revisits ZCS, a learning classifier system. The use of simple difference equation models of ZCS are shown to demonstrate how the system is capable of optimal performance subject to appropriate parameter settings. In contrast to current research trends emphasizing accuracy-based fitness, this study re-examines a payoff-based approach. Simple difference equation models of ZCS are used to show that the system is capable of optimal performance subject to appropriate parameter settings. This capability is demonstrated for both single- and multistep tasks. It is presented to support findings from the models in well-known multistep maze tasks. This research offers a compelling case for the effectiveness of payoff-based learning classifier systems. The models and maze tasks support its use, challenging conventional approaches. It demonstrates the value of ZCS in achieving optimal performance in complex learning scenarios.
Published in Evolutionary Computation, this research on the ZCS learning classifier system fits the journal's focus on computational intelligence and evolutionary algorithms. The application of difference equation models contributes to understanding and optimizing performance in machine learning, a key theme for the journal's audience.