The previous three sections described how an evaluator measures timeliness of detection, sensitivity, and false alarm rate for a detection algorithm with a fixed threshold. As with case detection algorithms, if the evaluator alters the algorithm's threshold level, the measurements all change.
When studying an outbreak detection algorithm, an evalua-tor also constructs an ROC curve to understand the tradeoff between its sensitivity and false alarm rate. To understand the relationship between its false alarm rate and timeliness, an evaluator plots timeliness against the false-alarm rate. This type of plot is called an activity monitoring operating characteristic (AMOC) curve (Fawcett and Provost, 1999). From the AMOC curve, an evaluator can easily see the tradeoff between the false alarm rate and timeliness. For example, in Figure 20.5, we see that if we were to increase the detection threshold to reduce the false alarm rate from four per year to two per year, the average day of detection would increase from 2.6 days to 4.4 days. The price we would pay for a lower rate of false alarms is a delay in detection of 1.8 days.
Was this article helpful?