Harnessing Social Media Sentiment for Predictive Insights into the Nigerian Presidential Election

Political events are heavily influenced by social media, shaping public opinion and actions. Sentiment analysis of social media content helps policymakers, campaign planners, and analysts understand voter sentiments for informed decision-making. This study performs a comprehensive comparative analysis of traditional machine learning models— Logistic Regression, Random Forest, Naïve Bayes, and SVM— and deep learning models—FFNN, LSTM, and CNN—on tweets collected via the X (formerly Twitter) API regarding the 2023 Nigerian Presidential Election. All models underwent a proper optimisation process and were evaluated using key performance evaluation metrics. Over 1.9 million tweets were collected over eight months. Results show deep learning models outperform traditional ones, with LSTM achieving the highest accuracy (95%), followed by CNN (94%) and FFNN (94%).