2. Objectives
• Predict vote cast for each Party in the 2018 Elections
in Minnesota
• 8 United States House of Representatives seats in
Minnesota – one from each of the 8 Congressional
Districts
• 2 United States Senate seats in Minnesota
• The Minnesota Governor Race
• Overall Predictions:
3 parties * 8 US House Seats +
3 parties * 2 US Senate Seats +
3 parties * 1 MN Governor Race
Objectives
APPROACH: Y=total voting Regression
3. Data Processing
Non-separatedSeparate by district Non-separated
2000~2016, House of Representatives
• Separate by congressional district
• Identify & adopt similar districts,
based on demographic, geographic
& political characteristics
2006~2016, Governor & Senate
• Include similar states into dataset to
train model, based on demographic,
geographic & political
characteristics
Iowa
Wisconsin
Michigan
4. Data Processing
Iowa
Anomaly Detection – the third parties matter
Basic Data Cleaning, tidying & Transformation
• In 2008, there was no third-party
competing the senator seats in Iowa,
so there is huge bias that cannot be
detected by predictors
• In 2008, there was great number of
votes for three third-parties,
especially for Independence. Since
there’s only one third party this time
and is not strong third party like
Independence, the figure for this year
is smoothed by average ratio.
9. • Web scraping Twitter October posts on annual base,
generated an average of 31,000 posts/per party per
election year.
• Apply sentimental analysis score on each post, then
average overall scores for each party per year
• Each sentimental figure is weighted by the professional
level of the user.
Social Media
Traffic
Feature Concentration