Transformer-based Framework for Election Outcome Prediction using Electorate Opinion

The accurate prediction of election outcomes plays a pivotal role in enhancing democratic processes by offering insights into voter behaviour and informing strategic decision-making for policymakers, political parties, and media organizations. This study introduces a novel framework leveraging the Robustly Optimized BERT Pretraining Approach (RoBERTa) to predict election results through sentiment analysis of public opinions expressed on X (formerly Twitter). By analyzing nuanced linguistic patterns in social media discourse, the research addresses the challenges of sentiment ambiguity and class imbalance. A case study on the 2023 Nigerian presidential election, focusing on Akwa Ibom State, demonstrates the effectiveness of this framework. The RoBERTa model achieved notable accuracy in predicting election outcomes, highlighting its potential for bridging the gap between online sentiment and real-world electoral results. The model achieved a higher precision, recall, and F1-score for the Negative class, with values of 0.95, 0.97, and 0.96, respectively, indicating exceptional performance in identifying and classifying negative sentiments. This framework underscores the transformative power of transformer-based architectures in electoral studies, offering avenues for more transparent and data-driven decision-making in political analysis.