The proper balance between sensitivity and FA rate is often application specific. The seizure detection problem for critical care applications requires a low false alarm rate. For this competition we measure senstitivity and false alarms using a Time-Aligned Event Scoring (TAES) metric that considers the percentage overlap between the two events and weights errors accordingly based on the amount of overlap.

These scores were then combined into a single scalar scoring metric:
P = sensitivity – 2.5 * (false alarm rate/24 hrs) – 7.5 * (number of channels / 19)

This measure has been shown to correlate well with a system’s ability to accurately segment a signal into seizure and background events.