A comparative analysis of multiple ML models for fake news detection

Abstract

The rapid growth of internet usage has emerged as a defining trend in recent decades, profoundly influencing how individuals communicate, access information, and engage with digital platforms. Among these, social media has become particularly prominent, with users demonstrating increasing proficiency in leveraging multiple platforms for diverse purposes. Such web resources enable any citizen to become a publisher or a distributor of news. The people and the content are not verified on these networks. People sometimes use such media to spread misleading information. The geometric growth of disseminated false information has become a critical concern in modern culture, which largely depends on information. The proliferation of fake and misleading information raises serious problems for societal well-being, the democratic process, and communal perception. Researchers have responded by increasingly using machine learning (ML) algorithms to develop automated systems to detect fake news. Based on this inspiration, we propose a resilient and effective system capable of appropriately identifying and combating misinformation propagation. This paper considered eight ML models to identify and evaluate fake news on two real-world datasets acquired on Kaggle. We use the Term Frequency-Inverted Document Frequency (TF-IDF) method in feature extraction. The best performance was recorded in the Passive Aggressive Classifier (PAC) among the other eight models, with an average accuracy of 97.26%.