|
Activity Number:
|
318
|
|
Type:
|
Invited
|
|
Date/Time:
|
Tuesday, July 31, 2007 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Statisticians in Defense and National Security
|
| Abstract - #307786 |
|
Title:
|
Multivariate Health Surveillance Using a Network of Bayesian Data Fusion Models
|
|
Author(s):
|
Zaruhi R. Mnatsakanyan*+ and Howard S. Burkom
|
|
Companies:
|
Johns Hopkins University and Johns Hopkins University
|
|
Address:
|
11100 Johns Hopkins Road, Applied Physics Laboratory, Laurel, MD, 20723,
|
|
Keywords:
|
Bayesian Belief Networks ; Data Fusion ; Decision Support System ; Anomaly Detection ; False Positive Reduction ; Biosurveillance algorithms
|
|
Abstract:
|
The model is a network of distributed Bayesian Network (BN) for region-wide surveillance of respiratory outbreaks using multiple data sources. Each BN combines inputs of alerting algorithms and preprocessed data to estimate the probability of true health anomalies at reduced alert rates. Each also discriminates between seasonal influenza and other significant events. The BN structures are based on association of anomalies with different data types and correlation of those events based on the decision-making logic of domain experts. The model detected Maryland influenza outbreaks in 2003-04, 2004-05 and did not alert for the 2005-06 season, which saw no significant rise in flu. The network fusion yields epidemiological false alarm rates significantly below those of univariate temporal alerting algorithms of ESSENCE surveillance systems.
|
