Data fusion for biosurveillance – A short survey and preliminary results


  1. Ko, A.
  2. Jousselme, A.-L.
  3. Maupin, P.
Corporate Authors
Defence R&D Canada - Valcartier, Valcartier QUE (CAN)
The primary concern of a bio-surveillance system is to analyze and interpret data as they are collected and then decide whether further investigation is required. Decision makers need to know whether the data in the current test interval are sufficiently different from expected counts to cause an alert. Sources of information providing syndromic time series are typically over-the-counter (OTH) drug sales, Emergency Department (ED) visits, number of diagnosis of certain diseases as registered by different hospital services, school absenteeism, nursery hotline, etc. The deviation from normal behavior of these data streams is called an outbreak. This report proposes a state of the art review of techniques for anomaly detection in multiple data streams. The following topic are covered: Different types of outbreak (or abnormal behaviors), methods for data pre-processing and time series feature extraction (windowing, sampling, segmentation, normalization, transformation, smoothing, regression), basic out-break detection methodologies, combination methods, monitoring of multiple time series (e.g., multiple testing problem using statistical fusion techniques), non­ statistical decision fusion techniques (classifier-based fusion, contingency tables, tree­ based fusion). We also propose a post-processing scheme, an important aspect usually ignored in the current biosurveillance literature, and which is used to refine biosurveillance classification. Series of preliminary experiments o

Il y a un résumé en français ici.

Syndromic surveillance;Anomaly detection;Disease outbreak;Times series;Data streams;Multiple Hypotheses Testing;Feature extraction
Report Number
DRDC-VALCARTIER-TR-2013-469 — Technical Report
Date of publication
01 Dec 2013
Number of Pages
Electronic Document(PDF)

Permanent link

Document 1 of 1

Date modified: