Comparative Performance of Machine Learning Algorithms for Fake News Detection

Abstract

Automatic detection of fake news, which could negatively affect individuals and the society, is an emerging research area attracting global attention. The problem has been approached in this paper from Natural Language Processing and Machine Learning perspectives. The evaluation is carried out for three standard datasets with a novel set of features extracted from the headlines and the contents. Performances of seven machine learning algorithms in terms of accuracies and F1 scores are compared. Gradient Boosting outperformed other classifiers with mean accuracy of 88% and F1-Score of 0.91.