DECODING THE BALLOT: PREDICTING INDIAN GENERAL ELECTIONS WITH MACHINE LEARNING

This paper explores the complexities of predicting election outcomes in India, focusing on the winning party and the probability of incumbent reelection. Leveraging historical voting data and socio-economic indicators from the Socioeconomic High-resolution Rural-Urban Geographic (SHRUG) dataset and the Lok Dhaba database, the study employs advanced machine learning models to forecast electoral results. The main goal of this paper is to find these models ability to forecast the victorious party and determine the likelihood of reelection is the main goal. Several models, including Random Forest, Gradient Boosting, and Decision Tree, were assessed to meet these goals. With an accuracy of 99.89%, the Random Forest model outperformed the rest of them. This is because of its ensemble learning strategy, which lowers overfitting and increases predictive power. Additionally successful were the Decision Tree and Gradient Boosting models, which yielded accuracies of 98.75% and 99.78%, respectively. The study faced challenges such as computational complexity and potential bias introduced by the dataset, particularly due to the historical dominance of the Indian National Congress (INC) party. Despite these challenges, the models provided valuable insights into voter behaviour and electoral trends. The implications of this study are significant for political analysts and campaign strategists. Accurate predictions can guide the development of targeted campaign strategies and enhance understanding of electoral dynamics. Future research should address dataset biases and explore more efficient algorithms to improve the robustness and applicability of these predictions in real-world scenarios.