Biology Reference
In-Depth Information
300
250
200
150
Authentic
Mimic
Same Class = 16/30
Proportion Same = 0.53
T(normalized) = 0.28
p−val = 0.78
100
Same class
Different class
50
10
20
30
40
50
60
giMilVisit
Figure 2.2
A simple example of the
knn test for multivariate distribution equality, using only two series
and 5 points from authentic and mimic. Each point is labeled as whether it is authentic or
mimic; the three nearest neighbors are computed, and an arrow is drawn connecting each
point to its three neighbors. The line is black if they are both the same type of point; gray if
different. The number of neighbor links that are the same is summed, then normalized, and
finally tested. Here, there is not enough evidence to reject the null hypothesis, so we conclude
that the authentic and mimic distributions may be the same.
2.5 outbreak Signature Simulation
2.5.1 Overview
We also consider the simulation of outbreak signatures to be added to
the background data. In order to compare the performance of biosurveil-
lance algorithms in terms of true and false alert rates and timeliness, we
simulate not only background data but also outbreaks signatures that can
be injected into the background data. Because in biosurveillance, the nature
of the outbreak signatures is generally unspecified, algorithms are tested
across different types and sizes of outbreak signatures.
Most researchers evaluating algorithm performance have added simulated
outbreak signatures to authentic data. Many studies add a fixed number of
additional cases, a linearly growing number of cases, or an exponentially
growing number of cases to the authentic data (Goldenberg et al., 2002; Reis
et al., 2003; Reis and Mandl 2003; Mandl et al., 2004; Stoto et al., 2006) in order
to provide a variety of different possible outbreak signal shapes. Single-day
“spike” signals and multiday lognormal curves are also popular (Burkom
2003b; Burkom et al., 2007), as they have some epidemiological basis. Spike
