A Robust Approach to Fake News Detection Using Dual Transformer Embeddings and Ensemble Classifiers

In the age of information, the rapid spread of fake news poses significant threats to public opinion and trust in media. The growing prevalence of fake news in digital media presents a serious threat to societal stability, public trust, and democratic discourse. Timely and accurate detection of fake news is thus critical for mitigating misinformation and maintaining the integrity of information ecosystems. This paper presents a robust and efficient fake news detection framework that combines the strengths of deep learning and ensemble learning techniques. The proposed method utilizes BERT and RoBERTa transformer models for extracting meaningful contextual embeddings from text data, which are then fused and classified using an ensemble of machine learning algorithms. The system was evaluated on a balanced dataset of real and fake news articles, achieving an accuracy of 97.85%, demonstrating superior performance over several existing approaches. The model's effectiveness was further validated on unseen data to simulate real world deployment scenarios. The results confirm the potential of the proposed method to serve as a reliable tool for automated fake news detection in modern digital ecosystems.