WSARE Conclusions

The WSARE algorithms approach the problem of early outbreak detection on multivariate surveillance data using two key components. In WSARE 2.0, the main component is an association rule search, which is used to find anomalous patterns between a recent data set and a baseline data set.

False Positives per Month figure 15.6 Asymptotic behavior of algorithms for simulated data.

The contribution of this rule search is best seen by considering the alternate approach of monitoring a univariate signal. If an attribute or combination of attributes is known to be an effective signal for the presence of a certain disease, then a univariate detector that monitors this signal will be an effective early warning system for that specific disease. However, if such a signal is not known a priori, then the association rule search will determine which attributes are of interest. We intend WSARE to be a general purpose safety net to be used in combination with a suite of specific disease detectors. The key to this safety net is to perform nonspecific disease detection and notice any unexpected patterns.

0 0

Post a comment