A Hybrid Abnormal Advertising Traffic Detection Method
Kun Wang, Guohai Xu, Chengyu Wang, Xiaofeng He · 2017
Abnormal traffic is pervasive in the online advertising market. There are various cheating approaches while traditional anti-fraud methods are only effective for specific patterns. Combining the rule-based methods with supervised classification methods, we propose an abnormal traffic detection framework on both user layer and traffic layer. On the user layer, rule-based filters are designed to detect malicious users with duplicate clicks. We extract hybrid features under multi-granular time windows and train a user classifier to filter cheaters and complex spams indirectly. On traffic layer, we apply traffic filters to detect explicit fraudulent clicks and use a prediction model to detect malicious traffic with a higher precision. Extensive experiments on ground-truth data demonstrate the effectiveness of our detection method.