Performance Analysis of the ML Prediction Models for the Detection of Sybil Accounts in an OSN

The Online Social Networks (OSNs) as such have significantly become huge platforms for information sharing and social interactions for a variety of users across the globe. In the backdrop of fast transformations, these OSNs are undergoing illegal activities especially in the form of security attacks, and have already started reflecting serious harmful effects on these interactions. One of the prominent attacks in such environments, the Sybil attack, is jeopardizing various categories of social interactions as the number of users having Sybil accounts on these social platforms is experiencing phenomenal growth. The existence of such Sybil accounts on OSNs may threaten to defeat the very purpose of these OSNs. The presence of these Sybil accounts of malicious users is really almost impossible to control and, very difficult to detect. In this paper, with the help of Machine Learning (ML), an attempt has been made to uncover the presence of such Sybil accounts on an OSN such as Twitter. After the acquisition and preprocessing of available datasets, the Correlation with Heatmap and Logistic Regression-Recursive Feature Elimination (LR-RFE) feature selection techniques were applied to get a set of optimal features from these datasets. Then the prediction models were trained on these datasets by using Random Forest (RF), Decision Tree (DT), Logistic Regression (LR) and Support Vector Machine (SVM) classifiers. Further, the effects of biasing of genuine accounts with fake accounts on feature selection and classification have been presented. It is concluded that the prediction models using the DT algorithm outperformed all other classifiers.