Malicious Account Detection on Indonesian Twitter Account

Abstract

The rapid growth of social networks indeed triggers an increase in malicious activities, including the spread of false information, the creation of fake accounts, spamming, and malware distribution. However, developing a detection system that can identify accounts precisely becomes quite challenging. In this paper, we present a study related to the detection of malicious accounts on Twitter users from Indonesia. Our study objective is to propose a simple feature set to detect malicious accounts using only a few metadata and the tweet content itself from Twitter. We divided the classification level into three: account level classification, tweet level classification, and combination of account and tweet level classification. To get the classification results, we applied some popular machine learning algorithms such as Random Forest, Decision Tree, AdaBoost Classifier, Neural Network, and Logistic Regression to each classification level. The results show that Random Forest achieved high classification accuracy (AUC >80%) in each classification level using our proposed feature set.