Misinformation on Twitter is a common issue, and its ill effect is endangering individuals, communities, and society. Its detection through efficient means will minimize its damage. Hence, we conducted a comparative analysis using different machine learning models and neural networks for Twitter misinformation detection with the development environment established in Google Colab and Python as the underlying programming language. We used a set of 44,898 labeled tweets drawn from Kaggle and labeled as fake or true news, and with four significant features. We tested and trained four models: Support Vector Machine (SVM), Recurrent Neural Network (RNN), Naive Bayes, and Random Forest (RF). We tested model performance through several metrics. The accuracy was 0.99 for SVM (precision: 0.99, recall: 0.99, F1-score: 0.99) and Random Forest (accuracy: 0.99, precision: 0.93, recall: 0.95, F1-score: 0.94), which was the highest among all other models. Naive Bayes (accuracy: 0.94, precision: 0.93, recall: 0.95, F1-score: 0.94) and RNN (accuracy: 0.79, precision: 0.94, recall: 0.64, F1-score: 0.76) performed relatively low. This study has contributed to the creation of strong misinformation detection systems, maintaining the integrity of online information and preventing the dissemination of false information on Twitter. Our comparative analysis sheds light on the most effective machine-learning methods for detecting misinformation, opening the door to future research and applications.
Ugwunna et al. (Wed,) studied this question.