Bucak, Serhat; Paulino, Alessandra; Pocklington, Richard

Brief description and goals

Our aim in this project was to implement two important data mining tasks on these topics: multi-class multi-label classification on bill categorization and multi-class single-class classification for vote prediction. Vote prediction, which is in fact a challenging task and has not been tried previously, is a more sophisticated way of analyzing and reasoning the behaviors of the senators. To achieve this goal, senator demographics, keywords for each bill that is accepted by the senate between the years 2005 and 2008 were collected from internet. In addition to the successful results in bill categorization task, we obtained encouraging results for vote prediction.   

Visualization:

By Party

The visualization by party consists of a prediction of the votes based on party and bill topic. The bill topics were classified using SVM and the vote prediction was generated by JRIPPER classifier.

By State

In this visualization, the user can see how likely is a state to vote NO for a particular topic, and compare with all the other states. Another option is to see some of the rules generated.