Article 9216

Title of the article

COMPARING THE POWER OF CHI-SQUARED AND CRAMÉR–VON MISES
CRITERIA FOR SMALL TEST SAMPLES OF BIOMETRIC DATA

Authors

Ivanov Aleksandr Ivanovich, doctor of technical sciences, associate professor, head of biometric and neuronal nets technology laboratory, Penza Scientific Research Electrotechnical Institute (440000, 9 Sovetskaja street, Penza, Russia), ivan@pniei.penza.ru
Gazin Aleksey Ivanovich, candidate of technical sciences, associate professor,sub-department of computer sciences, information technology and information protection, Lipetsk State Pedagogical University (398020, 42 Lenina street, Lipetsk, Russia), yearn@bk.ru
Vjatchanin Sergej Evgen'evich, associate professor, head of sub-department of radio and space communications, Penza State University (440026, 40 Krasnaya street, Penza, Russia), fvopgu@pnzgu.ru
Perfilov Konstantin Aleksandrovich, postgraduate student, Penza State University (440026, 40 Krasnaya street, Penza, Russia), fvopgu@pnzgu.ru

Index UDK

519.24; 57.017

Abstract

Background. Classical statistical criteria chisquare works bad with small text samples. This article is about of researching statistical criteria of Cramer – von Mizes works with small text samples. The propose of this research is compare of applicability statistical criteria of Cramer – von Mizes and statistical criteria chisquare with small text samples.
Materials and methods. It’s suggested to compare the power of this methods in the point of equiprobable type one and type two errors. It’s proved that in logarithmic scale equal probability errors of the power criteria of Cramer – von Mizes is straight line, this makes calculations easier.
Results and conclusions. Using the criterion of the Cramer-von Mizes instead of the chi-square test on a sample of 20 examples to reduce the likelihood of decisions made errors in one and a half times. If we consider the criteria compared some nonlinear low-frequency digital filters, the criterion of Cramer-von Mizes is more effective then chi-square due to the fact that there is a higher frequency and lower amplitude data quantization noise spikes.

Key words

Cramer – von Mizes statistical criteria, chi square criteria of Pearson, noise elimination of quantization.

Download PDF

 

Дата создания: 27.09.2016 14:32
Дата обновления: 29.09.2016 15:02