CONSTRAINED MIXTURE-OF-MIXTURE MODEL WITH APPLICATION TO KEYSTROKE DYNAMICS

• 成果类型: Article
• 署名作者: Simpson, Andrew; Michael, Semhar
• 署名单位: South Dakota State University
• 刊物名称: ANNALS OF APPLIED STATISTICS
• ISSN/ISSBN: 1932-6157; 1941-7330
• DOI: 10.1214/25-AOAS21042025
• 发表日期: 2025-12
• 页码: 3351-3371
• 关键词: Mixture of mixture model; hierarchical random effects model; keystroke dynamics; source identification; system security; constrained mixture model; em algorithm; likelihood; components

摘要

Keystroke dynamics has been used to both authenticate users of computer systems and to detect unauthorized users attempting to gain access. Many existing methods are supervised, assuming the true user of each keystroke is known a priori. However, this assumption does not always hold, particularly in businesses and government agencies that have internal systems accessible by multiple individuals. In such settings, unsupervised methods must be employed. One natural approach is to use finite mixture models to model keystroke dynamics. However, there is often not a one-to-one relationship between mixture components and users. Furthermore, users may type numerous times during a session or a block of time. In this case, the keystrokes from the session can be assumed to have originated from the same user. We propose a constrained mixture-of-mixture model that accounts for the lack of a one-to-one relationship between users and mixture components and the session-based grouping of keystrokes. Based on simulation studies and the motivating CMU keystroke dataset, the proposed model shows strong performance in an unsupervised setting of keystroke dynamics analysis.