A COMPREHENSIVE SURVEY ON MACHINE LEARNING TECHNIQUES FOR FRAUD DETECTION IN SOCIAL NETWORKS

Abstract

The proliferation of social media platforms has transformed how individuals communicate, disseminate material, and consume information. Moreover, social networks have evolved into a refuge for fraudulent activities such as automated bots, phishing schemes, disinformation operations, and counterfeit profiles. Machine learning methodologies have increasingly been employed for fraud detection to mitigate these emerging concerns. This paper provides a comprehensive examination of various machine learning models utilized in social network fraud detection, including traditional methods such as supervised and unsupervised learning, as well as advanced approaches like deep learning, ensemble techniques, and graph-based models. Furthermore, it emphasizes multimodal strategies that integrate social network frameworks, textual information, visual media, and user actions for enhanced detection accuracy. The study provides a vital evaluation of current methodologies regarding accuracy, scalability, and relevance to practical situations. The discussion encompasses essential datasets, assessment measures, and prevalent limitations within the existing research environment. The survey continues by noting existing shortcomings and emphasizing future research opportunities, such as interpretable models, resilient real-time systems, and privacy-preserving frameworks.