A cloud endpoint coordinating CAPTCHA based on multi-view stacking ensemble. Issue 103 (April 2021)

• Record Type: Journal Article
• Authors: Ouyang, Zhiyou; Zhai, Xu; Wu, Jinran; Yang, Jian; Yue, Dong; Dou, Chunxia; Zhang, Tengfei
• Is Part Of: Computers & security. Issue 103(2021)
• Publication Date: 2021-04
• Subjects: Anomaly detection -- Semi-supervised learning -- Ensemble Learning -- CAPTCHA -- Network Security,; Computer security -- Periodicals; Electronic data processing departments -- Security measures -- Periodicals; 005.805
• DOI: 10.1016/j.cose.2021.102178
• Languages: English
• ISSNs: 0167-4048

Abstract

Abstract: Fully Autonomous Public Turing test to tell Computers and Humans Apart (CAPTCHA) is an essential component for network security resisting attacks, such as collision attack and password blasting.As a recently emerged CAPTCHA technology, drag-and-drop interactive CAPTCHA has been successfully employed in great number of practical applications. However, there are still some problems involved in the architecture and back-end anomaly detection model of the interactive CAPTCHA that need to be addressed: excessive concentration of computing pressure on cloud system, poor accuracy of anomaly detection model, and huge cost of the labelling for the attack sample. To this end, a novel cloud endpoint coordinating CAPTCHA based on multi-view stacking ensemble (MVSE) is proposed in this paper. In particular, a novel cloud endpoint coordinating CAPTCHA architecture is designed to make most use of the computing power of endpoint devices and reduce the calculation pressure of cloud system. Meanwhile, a multi-view stacking ensemble learning-based user action anomaly detection model is proposed for the cloud endpoint coordinating CAPTCHA architecture. Finally, an iterative top-k training (ITK-training) semi-supervised learning algorithm is employed for data enhancement and make the most use of un-labeled samples in order to reduce the deploy cost of drag-and-drop CAPTCHA system. A real-world data from one of the biggest Internet companies of China is used to validate the effectiveness of our proposed model. We can obtain that the computing pressure of the cloud can reduce nearly 95% and the accuracy of the proposed CAPTCHA system can reach 96.77% using MVSE learning and 98.67% using MVSE learning with the ITK-training based data enhancement.