Political Vote Prediction

We developed a SVM-based machine learning system that correctly predicted the votes of the United States House of Respresentatives 84.60% of the time. A separate model is trained for each representative based on their past history of voting and pulling information such as party allegance, who's sponsoring a given bill, bag of words on the bill description, and other infomation. Each representative model is then used in combination to predict the final success of new bill.

For those representatives who voted along party lines, our accuracy was not much better than simply classifying based on party. However, for less strict representatives our system performed much better than the simple party-lines baseline.

The project was completed with Ben Shulmann and Jisha Kambo, with the advising of Igor Labutov and Thorsten Joachims

The final project was presented in a small showcase. If you're interested, feel free to check out the report or the poster.