SoftED: Metrics for soft evaluation of time series event detection

Abstract

Time series event detectors are evaluated mainly by standard classification metrics, focusing solely on detection accuracy. However, inaccuracy in detecting an event can often result from its preceding or delayed effects reflected in neighboring detections. These detections are valuable to trigger necessary actions or help mitigate unwelcome consequences. In this context, current metrics are insufficient and inadequate for the context of event detection. There is a demand for metrics that incorporate both the concept of time and temporal tolerance for neighboring detections. Inspired by fuzzy sets, this paper introduces SoftED metrics, a new set designed for soft evaluating event detectors. They enable the evaluation of the detection accuracy and the degree to which their detections represent events. A new general protocol inspired by competency questions is also introduced to evaluate temporal tolerant metrics for event detection. The SoftED metrics can improve event detection evaluations by associating events and their representative detections, incorporating temporal tolerance in over 36% of the overall detector evaluations compared to the usual classification metrics. Following the proposed evaluation protocol, SoftED metrics were evaluated by domain specialists who indicated their contribution to detection evaluation and method selection.