← Back to all projects
Of the People, By the People, For the People: Predicting US Presidential Polls per State using Machine Learning

Of the People, By the People, For the People: Predicting US Presidential Polls per State using Machine Learning

Jason Dolorso

Asian Institute of Management

Executive Summary

FiveThirtyEight, a leading poll analysis website, released national polling averages ahead of the 2020 US presidential election. What sets FiveThirtyEight’s forecast apart is its use of aggregated polling data from other firms and organizations, and its incorporation of COVID-19 into the model.

This study uses features such as demographics, unemployment rate, approval rating, and previous election results to build (1) a classifier to predict the outcome, and (2) a regressor to predict each candidate’s percentages per state, based on FiveThirtyEight’s polling averages.

Key Highlights

  • US states are likely to vote for the same party they voted for in 2016.
  • The current administration’s approval ratings play a big role in predicting poll results.
  • States with lower median age lean towards Trump, while higher unemployment rate and Asian population lean towards Biden.

Full text article and source codes can be provided upon request.