Adaptation of the idea of concept drift to some behavioral biometrics: Preliminary studies. (March 2021)

• Record Type: Journal Article
• Authors: Porwik, Piotr; Doroz, Rafal
• Is Part Of: Engineering applications of artificial intelligence. Volume 99(2021)
• Publication Date: 2021-03
• Subjects: Concept drift -- Biometrics -- Classifiers -- Ensemble of classifiers; Engineering -- Data processing -- Periodicals; Artificial intelligence -- Periodicals; Expert systems (Computer science) -- Periodicals; Ingénierie -- Informatique -- Périodiques; Intelligence artificielle -- Périodiques; Systèmes experts (Informatique) -- Périodiques; Artificial intelligence; Engineering -- Data processing; Expert systems (Computer science); Periodicals; 620.00285
• DOI: 10.1016/j.engappai.2020.104135
• Languages: English
• ISSNs: 0952-1976

Abstract

Abstract: In this paper we present a novel strategy that utilizes concept drift to improve some biometric procedures. The proposed method can be applied whenever behavioral signals change and those changes need to be detected. From a security point of view, this is important because detection of and appropriate response to change should result in some alteration in the operation of the biometric system. As one example, this allows for the detection of legitimate and illegitimate users. Experiments performed on real biometric signals have demonstrated that the proposed techniques could be introduced into existing professional biometric systems based on behavioral features. Highlights: We demonstrate that concept drift idea can be adapted to biometric domain. We showed that biometric behaviors can be monitored. It was confirmed by the Bayesian sign-rank test. We report that results deliver advantages. The ACC is higher and the FAR/FRR coefficients are lower. Based on the modified Kolmogorov–Smirnov test also outliers in data can be discovered. The solution can be run in a parallel environment. This significantly reduces the computation time.