Abstract
Online Tracking of Spatiotemporal Event Patterns (OTSEP) is important in the fields of smart home and Internet of Things (IoT), but difficult to be resolved due to various noises. On account of the strong learning capability in noisy environments, Learning Automaton (LA) has been adopted in the existing literature to notify users once a pattern disappears, and suppress the notification to avoid the distraction from noise if a pattern exists. However, the LA-based models require continuous and identical responses from the environment to jump to another action, which lowers their learning speed especially when the noise level is high. This paper proposes a sliding window method, with which the learning speed is stable in different environments. Experimental results show that the learning accuracy and speed are greatly improved over the existing methods in dynamic and noisy environments.