In the current age of digital technology, social media now represents the main news source choice for consumers. This practical approach makes false information spread wider. As social media developed with media-rich platforms people shifted from reading fake news in single-text formats to consuming it among multiple media components. Considering that digital media have lots of advantages in the news industry and the Absence of protocols for control and verification, the spread of false news is among the greatest challenges to the economy, democracy, media, freedom of speech, and health. Thus, finding and utilizing efficient automated techniques for the detection of fake information on social media has turned out to be a major task. However, this topic has huge obstacles, the major one being a lack of resources, such as evaluation tools, computer power, and access to large datasets. The approaches currently under investigation by researchers to solutions to these problems are numerous. In this respect, we proposed a comparative analysis of deep learning architectures for identifying false information. We used transformer-based models BERT along with other traditional machine learning models. With an accuracy of 99.26% in detecting fake news, experimental results on actual datasets show that the suggested strategy, in particular, BERT, works better than previous approaches.