Leveraging Pre-trained Transformer Models and Ensemble Learning for Fake News Detection: A Comparative Analysis

In recent years, the proliferation of fake news has become a major concern due to its potential to influence public opinion, cause panic, and even impact election outcomes. To address this issue, Deep Learning and Natural Language Processing (NLP) techniques have been proposed as effective solutions. In this study, we present a comprehensive evaluation of transformer-based models, particularly BERT, DistilBERT, and RoBERTa, alongside ensemble learning techniques, to enhance the performance of fake news detection tasks. Specifically, we fine-tuned these pre-trained models on two widely used datasets: LIAR and ISOT, which represent different levels of complexity in fake news classification. Subsequently, our approach integrates individual transformer models with ensembling methods such as Majority vote, Averaging and Highest-sum, significantly improving classification accuracy. Extensive experiments have been carried out showing that the proposed approach achieves high accuracy in detecting fake news across various standard evaluation metrics, outperforming competitive state-of-the-art baselines.