Abstract:
This paper presents a machine learning-based approach for detection of malicious users in the largest Russian online social network VKontakte. An exploratory data analysis was conducted to determine the insights and anomalies in a dataset consisted of 42394 malicious and 241035 genuine accounts. Furthermore, a tool for automated collection of the information about malicious accounts in the VKontakte online social network was developed and used for the dataset collection, described in this research. A baseline feature engineering was conducted and the CatBoost classifier was used to build a classification model. The results showed that this model can identify malicious users with an overall 0.91 AUC-score validated with 4-folds cross-validation approach.
Keywords:VKontakte, malicious users, machine learning, social networks, classification models.