Abstract
Users often observe anomalous behaviors of systems, such as machine failures, autonomous agents, and natural phenomena. We analyze the features and the benefits of the memory-based strategy, which focuses on memorization of instances to predict anomalous and regular behaviors of the system. In this study, we develop our previous research and investigate the cognitive processes and the benefits of the memory-based strategy with ACT-R model simulations. We set the parameters defining the encoding processes of anomalous instances and regular instances in the model of the memory-based strategy and performed simulations to verify how these two parameters influence prediction performance. The results of simulations showed that (1) anomalous instances are encoded and regular instances are not encoded in the memory-based strategy and that (2) such inactivity on regular instances suppresses commission errors of regular instances and does not suppress commission errors of anomalous instances and omission errors, which leads to correct prediction of systems' behaviors.
