Turning Data into Democracy:
AKA: Complex Political Phenomenon
Setting the Stage
In the ever-shifting landscape of politics, predicting the outcome of an election is no easy feat. Historically, analysts and pundits have relied on a combination of polling data, demographics, and intuition to forecast electoral outcomes. However, as election dynamics become more complex, traditional methods often fall short.
The question, then, is how can we improve the accuracy of our predictions? One promising approach lies in leveraging the vast wealth of historical polling data that we have accumulated over the years. By using this data to train machine learning models, we may be able to unlock new insights and improve the accuracy of our election predictions.
In this presentation, we will explore the process and challenges of harnessing historical polling data to create a predictive model for future elections.
TABLE OF CONTENTS
01
03
02
04
Stephen Whitson
Data Collection
Emmanuel Montano
Data Cleaning
Michael Schell
Programming Model
Robert Lehr
Visualizations
Data Collection
01
Stephen Whitson
Data Collection
Polling Data
The dataset for this project, sourced from Kaggle.com and fivethirtyeight.com and is a compilation of historical polling data from various organizations and research groups.
Data Cleaning
02
Emmanuel Montano
Data Cleaning
Data used
For data cleaning, we used pandas to load the data from SQLite database files, filtered it down to four columns, and exported the cleaned data as CSV and JSON files. We compiled the data into an SQLite database, chosen for its serverless hosting advantage. Our model training utilized data from the 2016 and 2019 elections. With our testing using the 2023 data.
Data Cleaning
Programming Model
03
Michael Schell
Programming Model
Look at the data used
Programming Model
Initial Random Forest Predictions
Random Forest Predictions
Results: .99
Interpretation: Trump has a 99% chance of winning.
Programming Model
XG Boost Training Model
XGBoost Predicting Model
Results: .23
Interpretation: Biden 77% of Winning.
Visualizations
04
Robert Lehr
Visualizations
Visualizations
Summary
A Complex-Political-Phenomenon
The M.E.R.S. Research Facility uses massive multivariate optimizing algorithmic models (MMOAMs), a.k.a. machine learning algorithms, to predict the winner of the next USA Presidential Election. The model predicts that there is a 77.4% that Democrats will win the Presidential election. Our model predicts with with 82.5% recall (i.e. - 17.5% the model inaccurately states Republican favorability is higher) and 100% precision (i.e, - 100%, the model accurately states Democrat favorability). Read more to find out how!
Any Questions?