Detection of Fake Accounts on Social Media using Machine Learning Techniques

Abstract

The rise of fictitious social media accounts has proven detrimental to user privacy, information authenticity, and public dialogue. Fake accounts, created and managed by bots or bad actors, are easily used for spreading false information, influencing and deceiving people, or committing other scams. This paper focuses on the use of machine learning approaches on the problem of detecting fake accounts on social media. We examine posting behaviours, language use, followed and following user identity ratios, and several metrics of network connectivity as both content and behaviour features. Ensemble learning, deep learning, and graph-based techniques like Graph Neural Networks (GNNs) are employed alongside supervised techniques like Support Vector Machines (SVM), Random Forests, and Logistic Regression. Ensemble and graph-based model results proved to be more robust and accurate than traditional classifiers. This research illustrates the need for multi-criteria and sophisticated learning approaches, which greatly facilitates scalable detection of fake accounts in uncontrollable social media settings.