Exploring social and economic predictors for U.S. Government elections



Journal Title

Journal ISSN

Volume Title



In democracy, elections are the way we settle our differences and redistribute power. It is the best assurance for avoiding the concentration of power by few and the shuffle of governments priorities. The main purpose of my research study is to develop a method to predict US government elections (US Presidential election, US Senate election and Election of US House of Representatives) outcome ahead of the time without using polling data. Thus, this study is focused on developing an alternative prediction tool that looks at the relations between a variety of variables and indicators that are associated with voting, among them historical, economic, and social indicators. On average when considering these different types of elections, the data available for each election are different. Therefore, it had to assume that for a given state voter turnout rate would be the same for each congressional district in that state. If those separate voter turnout data was available for the study, it would be greatly beneficial in predicting the House election. If the model incorporates demographic data such as, the population, education level, composition of ethnicity, income levels for a given state or better for a given congressional district, that will play a huge factor in predicting an outcome for an election. The mathematical model includes a novel competitive analysis involving the con- current use of Linear Discriminant Analysis (LDA), Support Vector Machine (SVM), and Long Short-Term Memory Neural Network (LSTM) learning models for each state individually, in search for the minimal forecasting error over the available elec- tion history and the economic factors. Drawing on advancements in mathematical modeling and artificial intelligence we were able to run thousands of simulations and generate predictions. Testing model, this tool was able to predict the outcome of 2020 Presidential, Senate and House Elections. The developed model was validated by using the past US Presidential elections, which yielded results with high accuracy rates. It was founded that the Voter turnout rates for elections has a significant impact on the outcome of an election. Therefore, according to different levels voter turnout rates several predictions were made for each type of election.



Election, Prediction, Senate, House, Presidential, Machine Learning