A Comparative Analysis of Ensemble Learning Methods on Social Media Account Detection

Abstract

Today, social media platforms usage and benefiting rate from these environments are increasing. This rapid spread of social media has also allowed the emergence of fake accounts. Fake accounts are generally created to implement malicious activities through another user account or to spread incorrect information. To prevent the detriment that this situation may cause to real individuals, an effective fake account detection was carried out by using ensemble learning methods (Bagging, Boosting, Stacking, Voting and Blending) in this study. These methods were combined with various machine learning algorithms to measure their effectiveness in detecting fake accounts. The experimental results suggested that Bagging technique attained an accuracy level of 90.441%, Stacking technique 89.706%, Voting technique 88.971% and the Blending technique attained 88.235% in the test phase. While for the Boosting methods, XGboost technique attained accuracy level of 86.765%, whereas the AdaBoost outperformed it with an accuracy level of 91.912% in the test phase. The extant results demonstrates that ensemble learning methods combined with machine learning algorithms are efficient in detecting fake social media accounts. It is considered that additional studies with larger datasets alongside the usage of different ensemble methods can further improve the accuracy of the detection process.