Estimating Accuracy of Mobile-Masquerader Detection Using Worst-Case and Best-Case Scenario

详细信息    查看全文

• 作者：Oleksiy Mazhelis ; Seppo Puuronen ; Mika Raento
• 刊名：Lecture Notes in Computer Science
• 出版年：2006
• 出版时间：2006
• 年：2006
• 卷：4307
• 期：1
• 页码：302-321
• 全文大小：578 KB
• 刊物类别：Computer Science
• 刊物主题：Artificial Intelligence and Robotics
  Computer Communication Networks
  Software Engineering
  Data Encryption
  Database Management
  Computation by Abstract Devices
  Algorithm Analysis and Problem Complexity
  
• 出版者：Springer Berlin / Heidelberg
• ISSN：1611-3349

文摘

In order to resist an unauthorized use of the resources accessible through mobile terminals, masquerader detection means can be employed. In this paper, the problem of mobile-masquerader detection is approached as a classification problem, and the detection is performed by an ensemble of one-class classifiers. Each classifier compares a measure describing user behavior or environment with the profile accumulating the information about past behavior and environment. The accuracy of classification is empirically estimated by experimenting with a dataset describing the behavior and environment of two groups of mobile users, where the users within groups are affiliated with each other. It is assumed that users within a group have similarities in their behavior and environment and hence are more difficult to differentiate, as compared with distinguishing between the users of different groups. From the practical detection perspective, the former case corresponds to the “worst-case” scenario where the masquerader has a rich knowledge of the user behavior and environment and is able to mimic them, while the latter case corresponds to the “best-case” scenario, where the masquerader makes little or no attempt to mimic the behavior and environment of the user. The classification accuracies are also evaluated for different levels of false rejection errors. The obtained results indicate that, when smaller values of false rejection errors are required, ensembles of few best-performing classifiers are preferable, while a five-classifier ensemble achieves better accuracy when higher levels of false rejection errors are tolerated.