Proceedings of the Annual Meeting of the Cognitive Science Society

Users sometimes face anomalous behaviors of systems, such asmachine failures and autonomous agents. Predicting suchbehaviors of systems is difficult. We investigate the benefits ofthe memory-based strategy, which focuses on memorization ofinstances to predict anomalous and regular behaviors of thesystem, with ACT-R simulations with a cognitive model. Inthis study, we presumed the parameters defining the encodingprocesses on anomalous instances and regular instances in themodel of the memory-based strategy and performedsimulations to verify how these two parameters influenceprediction performance. The results of simulations showed that(1) regular instances are not encoded as default values in thememory-based strategy and that (2) such inactivity on regularinstances suppresses commission errors of regular instancesand does not suppress commission errors of anomalousinstances nor omission errors.