Attribute Selection and Intrusion Detection for Knowledge-based Authentication Systems

D. Nikolić, B. Vuksanović, and C. Nguyen (UK)


Machine learning, knowledge-based authentication, intrusion detection, expectation-maximisation algorithm


This paper proposes new methods for attribute selection and intrusion detection for knowledge-based authentication (KBA) systems. The authentication challenge is a random set of questions generated from user details stored in the data base of a particular organisation or institution. Attribute se lection is a critical step in the process of creating authentica tion challenges. Intrusion detection part of the system applies clustering methods for the analysis of authentication data in order to detect possible intrusion attempts in the system. First, this paper describes the data pre-processing and im portance sampling methods applied for the selection of at tributes used in the challenge question generation as well as methods for the final formulation and formatting of the challenge questions. Next, the paper describes the application of the Expectation-Maximisation (EM) algorithm in the in trusion detection tasks and presents some results on the arti ficially generated sets of authentication data. The next stages of this project, not discussed in this paper, involve the design of challenge question formats subsequent to selection of user attributes and the design of an unbiased system for collection of empirical data for evaluating and improving the intrusion detection algorithm.

Important Links:

Go Back