Detecting Bengali Fake News: A Comparative Study of CNN, BiLSTM, BiGRU, and Transformer Models

Fake news has become a major problem, particularly in areas with few linguistic resources, such as Bengali populations. False information has the potential to seriously affect society and politics, while also misleading and confusing others. Reliable methods to automatically identify false information are desperately needed to combat this. This work investigates several machine learning methods to tackle the problem of detecting fake news in Bengali, a low-resource language. We tested the performance of BiLSTM, BiGRU, CNN, and Transformer models on a curated Bengali fake news dataset, including strategies such as upsampling and data augmentation to address class imbalance. Our results reveal that the Transformer model beats previous techniques, with an accuracy of 99.98%. This study establishes a standard for detecting fake news in Bengali and illustrates the possibility of applying similar approaches to other underrepresented languages, helping the worldwide battle against disinformation.