Malicious domain detection based on k-means and smote

Abstract

The Domain Name System (DNS) as the foundation of Internet, has been widely used by cybercriminals. A lot of malicious domain detection methods have received significant success in the past decades. However, existing detection methods usually use classification-based and association-based representations, which are not capable of dealing with the imbalanced problem between malicious and benign domains. To solve the problem, we propose a novel domain detection system named KSDom. KSDom designs a data collector to collect a large number of DNS traffic data and rich external DNS-related data, then employs K-means and SMOTE method to handle the imbalanced data. Finally, KSDom uses Categorical Boosting (CatBoost) algorithm to identify malicious domains. Comprehensive experimental results clearly show the effectiveness of our KSDom system and prove its good robustness in imbalanced datasets with different ratios. KSDom still has high accuracy even in extremely imbalanced DNS traffic.