Detection of Fake Profiles on Online Social Network Platforms: Performance Evaluation of Artificial Intelligence Techniques

The emergence of online social network (OSN) platforms has resulted in the production of enormous amounts of data from billions of active users. The ease of access to personal information on OSN platforms renders users vulnerable to the creation of fake profiles. The primary objective of fake accounts is to disseminate unsolicited messages, unverified information, and other deceitful content on OSN platforms. No study has been conducted to ascertain whether traditional machine learning techniques or deep learning techniques are more effective at detecting fake profiles on OSN platforms with respect to dataset size. The present study fills this void by conducting a performance evaluation of artificial intelligence (traditional machine learning and deep learning) techniques using benchmark datasets of different sizes. An ablation study is conducted to ascertain the optimal combination of features, while data augmentation techniques are employed to address the issue of an imbalanced dataset. The study’s results show that using the data augmentation technique, particularly the synthetic minority over-sampling technique (SMOTE), yields better results on an imbalanced dataset. Further deep learning techniques (LSTM, which has an accuracy rate of 97%) work better on large datasets for finding fake profiles on OSN platforms, while traditional machine learning techniques (XGBoost, which also has an accuracy rate of 97%) work better on small datasets. The top-performing techniques are also compared with state-of-the-art techniques to validate the results. The study may aid future researchers in developing a comprehensive methodology for detecting fake profiles on OSN platforms.