Online Social Networks Flu Trend Tracker: A Novel Sensory Approach to Predict Flu Trends - Biomedical Engineering Systems and Technologies - page 342

The colormap used is a scale with white representing low intensity and black, high
intensity. We are comparing “trends” among the ILI and Twitter data.
Regional analysis shows that ILI seems to peak later in the Northeast (Regions 1
and 2) than in the rest of the country by at least week. The Twitter reports also follow
this trend. In Region 9, Region 4 and the Northeast, the ILI rates seem to drop off
fairly slowly in the weeks immediately following the peaks. This is also reflected in
the Twitter reports. Approximately 20-25 weeks after the peak ILI, the northern regions
have lower levels relative to the peaks in the southern regions. This is also true of the
Twitter reports. The decline in ILI rates is slowest in Region 9.
Figure 8 depicts regionwise ILI prediction performance for the year 2010-2011 using
our logit model. We select region 1, region 6 and region 9 to represent the regions, one
each from the East, South and Western U.S. and plot the actual and predicted ILI values
for each of these regions using Twitter data, Facebook data, and the combination of
Twitter and Facebook data. We observe that the OSN reports and ILI rates are in fact
correlated across regions and therefore corroborate our earlier findings that OSN can
improve ILI rate prediction.

7 Conclusions

In this paper, we have described our approach to achieve faster, near real time prediction
of the emergence and spread of influenza epidemic, through continuous tracking of flu
related OSN messages originating within United States. We showed that applying text
classification on the flu related messages significantly enhances the correlation between
the Twitter and Facebook data and the ILI rates from CDC.
For prediction, we build an auto-regression with exogenous input (ARX) model
where the ILI rate of previous weeks from CDC formed the autoregressive portion of
the model, and the OSN data served as an exogenous input. Our results indicated that
while previous ILI rates from CDC offered a realistic (but delayed) measure of a flu
epidemic, OSN data provided a real-time assessment of the current epidemic condition
and can be used to compensate for the lack of current ILI data.
We observed that the OSN data was highly correlated with the ILI rates across differ-
ent HHS regions. Therefore, flu trend tracking using OSN's significantly enhances public
health preparedness against the influenza epidemic and other large scale pandemics.

Acknowledgements. This research is supported in parts by the National Institutes of
Health under grant 1R43LM010766-01 and National Science Foundation under grant
CNS-0953620.

References

1. Achrekar, H., Gandhe, A., Lazarus, R., Yu, S.H., Liu, B.: Predicting flu trends using twitter
data. In: International Workshop on Cyber-Physical Networking Systems (June 2011)
2. Achrekar, H., Gandhe, A., Lazarus, R., Yu, S.H., Liu, B.: Twitter improves seasonal influenza
prediction. In: Fifth Annual International Conference on Health Informatics (February 2012)
3. Centers for Disease Control and Prevention: FluView, a weekly influenza surveillance report
(2009), http://www.cdc.gov/flu/weekly