Online event detection for sensor data

Resumo

In streaming time series analysis, it is often possible to observe the occurrence of a significant change in
behavior at a certain point or time interval. Such behavior change generally characterizes the occurrence of an
event. An event can represent a phenomenon with defined meaning in a domain of knowledge. The event detection
problem becomes particularly relevant in this context, especially for applications based on sensor data analysis.

The algorithms for detecting events online or in real-time run simultaneously with the process they are monitor-
ing, processing each data point as they become available. Online event detection for streaming applications is a

challenging problem that creates an increasing demand for high-performance computing and advanced machine

learning techniques. Although there is a wide variety of methods, no silver bullet technique exists for event detec-
tion. In this context, this work contributes by providing a taxonomy for online detection of events in time series,

including incremental and adaptive learning and some of the main methods addressed in the literature.