I used a machine learning model to predict the outcomes of the 2018 Midterm Congressional races. I did this by downloading 2018 census data for each congressional district. This ended up being around 12.7K census variables, so I used Principal Components Analysis to reduce those 12.7K features down to just 3 dimensions. The following plots show what my model predicted using those three dimensions. The markers are shaded dark blue if the model correctly predicted the district to go Democrat; and they are shaded dark red if my model correctly predicted the district to go Republican. They are shaded cyan if the model misclassified the district as being Republican, when in actuality it went Democrat in 2018. Likewise, it is shaded orange if the model misclassified a Republican district as being Democrat. In the first 4 plots, the third dimension is represented by the size of the marker. The 5th plot is a 3D plot and as such as a z axis. Each of the following plots is interactable with tooltips; the 5th plot (3D) can be panned and zoomed.
