Identifying Fraudulent Social Media Profiles: An ML and Text Analysis Framework

The proliferation of fraudulent social media profiles poses significant challenges for online security and trust. This paper presents a novel machine learning and text analysis framework that significantly improves fake profile detection by integrating linguistic patterns, behavioral metadata, and network features. Unlike existing approaches that rely on single detection methods, our hybrid model combines BERT-based text analysis, ensemble learning, and anomaly detection to achieve 94.2% accuracy across multiple platforms while maintaining robustness against adversarial evasion. A key innovation is the incorporation of explainable AI (XAI) techniques, which provide interpretable decision-making for human moderators. Our experiments demonstrate superior performance compared to state-of-the-art methods, particularly in detecting sophisticated fake profiles that evade traditional filters. The framework’s cross-platform generalizability and real-time applicability make it a practical solution for social media platforms combating fraudulent activity. These advancements address critical gaps in fake profile detection by improving accuracy, transparency, and adaptability—key requirements for modern content moderation systems.