Fake news is a lasting issue in our society due to their propagation speed and their impact on public opinion. In recent years, the internet and social media have offered a perfect ground for misinformation to spread, as people can comment, elaborate and share fake news without any control. Consequently, detecting and preventing the dissemination of fake information has indeed become a pressing policy issue to tackle. For this reason, in this paper we briefly review existing fake news datasets and we present an integrated statistical methodology for fake news detection, based on text mining and classification. An application on the ISOT Fake News dataset, regarding 2016 US presidential elections, shows that ensemble methods are the most reliable in classifying fake news articles from their textual content.