I've always had an interest in politics. So when I learned about supervised learning algorithims at Metis, in which you train the model to predict a binary classification, the first thing that came to mind was 'Repulican' and 'Democrat'. My project was inspired in part by the current presidential election, but also in part by election fatigue. I was curious whether any of the horse race politics matters, and if I could build a model that could predict how a county will vote in an election regardless of current events and purely based on the demographics of the county. My final model (a random forest classifier) predicted which way counties will vote in the 2008 presidential election with 80% accuracy.
For future work, I would like to assess whether changes in a county over a four year term will affect how a county will vote.