Forecasting Election Sentiments: Deep Learning vs. Traditional Models

Sentiment analysis plays a pivotal role in understanding public opinion, especially during election periods. This research paper presents a comprehensive investigation into predicting sentiment trends before polls by employing advanced machine learning techniques. Our study evaluates the efficacy of a deep neural network model in comparison to traditional models, namely Naive Bayes and Logistic Regression. We utilise a dataset containing corruption indices and popularity scores as features to assess sentiment. The deep neural network model exhibits promising results, achieving an accuracy of 79.00% in predicting sentiment. Further analysis reveals precision, recall, and F1-Score values of 57.94%, 62.63%, and 60.19%, respectively. These outcomes highlight the model's ability to capture intricate sentiment patterns. In contrast, the Naive Bayes model achieves an accuracy of 52.50%, while Logistic Regression attains 54.00%. Their corresponding precision, recall, and F1-Score values indicate moderate predictive capabilities. The results underscore the potential of deep neural networks in sentiment prediction before polls, offering improved accuracy and capturing nuanced sentiment expressions. The comparative analysis suggests the need for future optimization and refinement of all models for more robust predictions. Our findings contribute to enhancing sentiment analysis methodologies for electoral forecasting, providing valuable insights for political decision-making processes.