ecms_neu_mini.png

Digital Library

of the European Council for Modelling and Simulation

 

Title:

Statistical Classification In Monitoring Systems

Authors:

Alexander A. Grusho, Nick A. Grusho, Elena E. Timonina

Published in:

 

 

(2016).ECMS 2016 Proceedings edited by: Thorsen Claus, Frank Herrmann, Michael Manitz, Oliver Rose, European Council for Modeling and Simulation. doi:10.7148/2016

 

 

ISBN: 978-0-9932440-2-5

 

30th European Conference on Modelling and Simulation,

Regensburg Germany, May 31st – June 3rd, 2016

 

Citation format:

Alexander A. Grusho, Nick A. Grusho, Elena E. Timoninaj (2016). Statistical Classification In Monitoring Systems, ECMS 2016 Proceedings edited by: Thorsten Claus, Frank Herrmann, Michael Manitz, Oliver Rose  European Council for Modeling and Simulation. doi:10.7148/2016-0658

 

DOI:

http://dx.doi.org/10.7148/2016-0658

Abstract:

The paper is devoted to the statistical classification problems. Repeated classification in control and monitoring systems is complicated by nonzero mistakes of traditional statistical decisions. At repeated applications of rules of statistical classification small probabilities of mistakes generate a large number of wrong decisions. At construction of monitoring systems of information security in computer systems wrong decisions are especially dangerous. Therefore for construction of secure architecture of control and monitoring systems it is necessary to look for nonconventional statistical decisions.

In finite set of words of finite length the ban is the word having zero probability of appearance. If the statistical criterion has the critical set consisting of only bans of supposed probability measure, the probability of wrong rejection of this measure is equal to zero. Therefore repeated application of such criterion won’t generate to false alarms in monitoring systems.

In the paper we consider a case of statistical classification when classes are defined by finite sets of probability distributions on a space of infinite sequences. We use bans to define decision functions and prove conditions when these decisions produce no mistakes.

 

Full text: