Ahead of the 2024 US presidential election, researchers at University of Delhi's Faculty of Management Studies (FMS) predicted a popular vote share in favour of incumbent Democratic candidate Kamala Harris.
The prediction is the result of a suite of machine learning algorithms employed by the researchers.
The study, led by Professors Pankaj Sinha, Amit Kumar, Sumana Biswas, and Chirag Gupta, combines economic and social indicators with advanced machine learning techniques to arrive at a forecast for Harris's performance as the Democratic nominee.
The research identifies six key variables impacting voter behavior - Gallup approval ratings (June and average), scandal ratings, unemployment rate, oil prices, and crime rates.
Meanwhile, three distinct machine learning models - Lasso, Random Forest, and Gradient Boosting - were also utilised in order to process these variables, each bringing unique predictive strengths.
The Lasso model, which minimises overfitting by reducing less relevant variables, emerged as the most consistent performer, forecasting a 47.04 per cent popular vote share for Harris.
Backtesting of these algorithms on past election data from 2012, 2016 and 2020 showed that Lasso maintained low prediction error rates, providing the most reliable forecasts.
By contrast, the other models, Random Forest and Gradient Boosting, demonstrated high accuracy on the training dataset but fell short in consistency due to overfitting on limited data samples.
Highlighting the importance of predictive accuracy for geopolitical and economic forecasting, the FMS researchers emphasised that their model could offer valuable insights for political strategists and economic analysts worldwide.
Copyright©2024 Living Media India Limited. For reprint rights: Syndications Today