Identifying fake news in political discourse behavior using machine learning methods

The paper examines the current problem of the rapid spread of political fake news, which can significantly influence public opinion, electoral processes, and intergovernmental relations.It highlights how the growth of online content and the dominance of social media as a source of information are speeding up the spread of misinformation and making it harder to spot it in time.On a theoretical level, the paper is based on the classification of approaches to fake detection (linguistic, topic-agnostic, machine learning, knowledge-based, and hybrid), emphasizing that only 28% of studies today focus specifically on ML methods, which determines the scientific and practical relevance of the proposed approach.The BERT model demonstrated the highest quality (Accuracy = 0.997; F1 = 0.998; ROC AUC = 0.997), significantly outperforming Random Forest (0.909; 0.930; 0.892) and Naive Bayes (0.881; 0.905; 0.870).Thus, modern transformer architectures provide not only maximum accuracy but also a balance between the completeness and specificity of predictions.The results obtained create the basis for the formation of highly accurate systems to counter disinformation, capable of responding quickly to the challenges of the modern media space.