2016 Election Primaries Classfication Tree

The primaries provide an opportunity to look at how different regions across the country respond to each candidate.  Using open source data on election results and demographics of each participating primary county, I developed a multiclass classification using CART tree modeling.   Decision tree learning is one of the foundational techniques of predictive analytics. While they have progressed into more robust ensemble methods, they continue to be a key component of data mining.  


Our data consisted of over one-thousand counties and the winning candidate by vote percentage.  The tree structure is based on the explanatory importance of each demographic feature (left to right).  For example, nearly all counties with greater than 27% African-American population voted for Hillary Clinton.  This result gives us our first end node to Hillary Clinton.  True responses to each decision node will follow the higher path and false responses will follow the lower.  Click on each node to expand and see how each candidate's key demographics develops.

Visualizations by Spencer Davison and inspired by d3js.org, Tableau Public, bl.ocks.org and various other sites in the analytics community.