The Detection of Fake Messages using Machine Learning

This research investigates how fake messages are used on Twitter during the Dutch election of 2012. It researches the performance of 8 supervised Machine Learning classifiers on a Twitter dataset. We provide that the Decision Tree algorithm perform best on the used dataset, with an F-Score of 88%. In total, 613.033 tweets were classified, of which 328.897 were classified as true, and 284.136 tweets were classified as false. Through a qualitative content analysis of false tweets sent during the election, distinctive features and characteristics of false content have been found and grouped into six different categories.