Biomedical Engineering Reference
In-Depth Information
Ta b l e 2 . Correlation between OSN Datasets and CDC along with its Root Mean Square Errors
(RMSE)
Twitter Facebook
Syndrome Elapse Retweets Correlation RMSE Correlation RMSE
Time
coefficient
errors
coefficient
errors
0 week
No
0.8907
0.3796
0.8728
0.4287
1 week
No
0.8895
0.3818
0.8709
0.4314
2 week
No
0.8886
0.3834
0.8698
0.4332
3 week
No
0.886
0.3878
0.8689
0.4346
4 week
No
0.8814
0.3955
0.8681
0.4357
10 0
Fitted Line
CCDF
CCDF
Fitted Line
−3
10 −1
−4
−5
10 −2
−6
10 −3
−7
−8
10 −4
−9
−10
10 −5
−11
10 −6
−12
10 0
10 1
10 2
1
1.5
2
2.5
3
3.5
4
4.5
Number of Tweets x
Number of Posts x
Fig. 5. Complementary Cumulative Distribution function (CCDF) of the number of tweets/posts
on Twitter/Facebook by same users
Most of these high-volume tweets in Twitter are created by health related organi-
zation, who tweet multiple time during a day and users who subscribe to flu related
RSS feeds published by these organizations. “Flu alert”,“swine flu pro”, “live h1n1”,
“How To Tips”, “MedicalNews4U” are examples of such agencies on Twitter. Simi-
larly one can identify agencies like “Flu Trackers”, “Influenza Flu” and specific users
that actively post on Facebook.
5
Prediction Model
The correlation between OSN activity and CDC reports can change due to a number of
factors. Annual or seasonal changes in flu-related trends, for instance vaccination rates
that are affected by health cares, result in the need to constantly update parameters relat-
ing OSN activity and flu activity. However, particularly at the beginning of the influenza
season, when prediction is of most significance, enough data may not be available to
accurately perform these updates. Additionally predicting changes in ILI rates simply
due to changes in flu-related OSN activity can be risky due to transient changes, such
as changes in OSN activity due to flu-related news.
In order to establish a baseline for the ILI activity and to smooth out any unde-
sired transients, we propose the use of Logistic Autoregression with exogenous inputs
(ARX). Effectively, we attempt to predict a CDC ILI statistic during a certain week by
Search WWH ::




Custom Search