A Comparative Analysis of Machine Learning Classifiers for Fake News Detection using NLP

Abstract

The swift propagation of fake news through social media and other digital outlets represents a significant problem for information integrity and public trust in the media. We present a comparative study of machine learning models for detecting fake news in an automated fashion with a focus on a Logistic Regression, Decision Tree, Gradient Boosting and Random Forest classifiers. The paper details a straightforward methodology employing Natural Language Processing (NLP) approaches to preprocess and transform textual data: cleaning the text, removing stop words, and stemming before applying the Term Frequency - Inverse Document Frequency (TF-IDF) method for vectorization in machine learning models. Once trained on a balanced dataset of real and fake news articles, we then report comparative performance of these classifiers using key metrics including accuracy, precision, and recall, and show the results of distinguishing real from fake news. This paper details applied, interpretable, and scalable work to combat mis- and disinformation and fake news; and offer a foundation for future work or papers employing higher order techniques and datasets.