Advancing Fake News Detection: From Traditional Machine Learning to Transformer-Based Approaches

Fake news on online platforms is spreading at a rapid rate, posing great threats to social, political, and economic stability in linguistically diverse regions like Kerala, India. This paper reviews comprehensively the techniques of fake news detection from traditional machine learning methods to transformer-based models. Special attention is given to the detection of fake news in regional languages like Malayalam, characterized by complex morphology and syntax. The study investigates applications of ensemble-based learning as well as state-of-the-art transformer models, BERT, and XLNet, focusing on their ability to capture nuanced linguistic features and contextual relationships. Evaluation results show how significantly better the transformer-based models fare than traditional methods with respect to accuracy and robustness. This review also captures current limitations - such as resource constraints for regional languages - as well as future opportunities in developing scalable, multilingual fake news detection systems.