Ashok K. Singh (Hospitality), along with Noah Loewy a student at Schreiber High School in Port Washington, New York; and Tina Marie Gallagher, co-coordinator of math research at that same high school wrote a paper, “Developing an Empirical Model to Forecast United States Presidential Elections: A Machine Learning Approach,” which has been published in Advances in Social Sciences Research Journal. This paper develops two models for forecasting the 2020 United States presidential election using the statistical method of multiple linear regression analysis (MLR) and Machine Learning method of Extreme Gradient Boosting (XGBoost). Seven predictor variables from 1976-2016 are chosen, as well as dummy columns for state variables to predict the Republican vote share in each state in the 2020 election. To deal with prediction uncertainties, we compute confidence intervals for the predicted values. We observe that 1) the probability of the Republicans winning the Electoral College is 36.3% when using XGBoost, yet 82.1% with MLR, 2) this election will boil down to four states: Michigan, New Hampshire, Pennsylvania, and Wisconsin, and 3) gradient boosting is a viable alternative to MLR in election forecasting. The large discrepancy in Electoral College predictions can be attributed to the XGBoost’s wider confidence interval and ability to account for non-linear relationships among the variables. Both methods predicted the Democratic candidate to win.
