In this project we used Deep Learning and Random Forest Machine Learning approaches to analize the 2018 Congressional Election and make models that could be applicable to 2020
Neural Network and Random Forest
Our analysis employed Deep Learning in the form of a neural network constructed with the scikitlearn library. We brought in over 323 features from Federal Election Commission data on the financial strength of Congressional Campaigns and United States Census Bureau data on the social, economic, and demographic make-up of every Congressional District in the United States
We found our initial model to be extremely over-fit to the 2018 election and so in order to make our model more applicable to future elections and more predictive based on similar data in the future we used a Random Forest approach to determine the most salient/predictive factors then trimmed our neural network to features determined by that process, and were left with 43, and a much more applicable, less over-fit, model.
Future Work
Our project can be potentially expanded with more historical campaign finance data and historical census data to make it even more robust.