Accounting for the effect of correlations by modulo averaging as part of neural network integration of statistical tests for small samples

The Aim of the paper is to demonstrate the advantages of taking into consideration real correlations by means of their symmetrization, which is significantly better than completely ignoring real correlations in cases of statistical estimation using small samples. Methods. Instead of real correlation numbers different in sign and modulo, identical values of correlation numbers moduli are used. It is shown that the equivalence of transformation to symmetrization is subject to the condition of identical probabilities of errors of the first and second kind for asymmetrical and equivalent symmetrical correlation matrices. The authors examine the procedure of accurate calculation of equal data correlation coefficients by trial and error and procedure of approximate calculation of symmetrical coefficients by averaging the moduli of real correlation numbers of an asymmetrical matrix. Results. The paper notes a practically linear dependence of equal probabilities of errors of the first and second kind from the dimension of the symmetrized problem being solved under logarithmic scale of the variables taken into consideration. That ultimately allows performing the examined calculations in table form using low-bit, low-power, inexpensive microcontrollers. The examined transformations have a quadratic computational complexity and come down to using pre-constructed 8-bit binary tables that associate the expected probability of errors of the first and second kind with the parameter of equal correlation of data. All the table calculations are correct and do not accumulate input data round-off errors. Conclusions. The now widely practiced complete disregard of the correlations when performing statistical analysis is very detrimental. It would be more correct to replace the matrices of real correlation numbers with symmetrical equivalents. The approximation error caused by simple averaging of the moduli of coefficient of asymmetrical matrices decreases as the square of their dimension or the square of the number of neurons that generalize classical statistical tests. When 16 and more neurons are used, the approximation error becomes negligible and can be disregarded.

replacement of statistical test with equivalent neurons, multicriteria statistical analysis of small samples, accounting for the effect of correlations, symmetrization of correlation matrices

1. R 50.1.037-2002. Recommendations for standardization. Applied statistics. Rules of check of experimental and theoretical distribution of the consent. Part I. Chi-square criteria. Moscow: Gosstandart Rossii; 2001. (in Russ.)

2. Ivanov A.I., Kupriyanov E.N., Tureev S.V. Neural network integration of classical statistical tests for processing small samples of biometrics data. Dependability 2019;2:22-27. DOI: 10.21683/1729-2646-2019-19-2-22-27.

3. Akhmetov B.B., Ivanov A.I. Estimation of quality of a small sampling biometric data using a more efficient form of the chi-square test. Dependability 2016;16(2):43-48.

4. Volchikhin V.I., Ivanov A.I., Bezyaev A.V., Kupriyanov E.N. The Neural Network Analysis of Normality of Small Samples of Biometric Data through Using the Chi-Square Test and Anderson–Darling Criteria. Engineering Technologies and Systems. 2019;29(2):205-217. DOI: 10.15507/2658-4123.029/2019.02.205-217.

5. Ivanov A.I., Bannych A.G., Kupriyanov E.N. et al. Collection of artificial neuron equivalent statistical criteria for their use when testing the hypothesis of normality of small samples of biometric data. Proceedings of the I All-Russian Science and Technology Conference Security of Information Technology. Penza. 2019. 156-164.

6. Perfilov K.A. [Criterion of geometric mean used for validity verification of the statistical hypotheses of biometric data distribution]. Proceedings of the Science and Technology Conference of the Penza Information Technology Security Cluster. Penza. 2014;9:92-93. [accessed 14.04.2020]. Available at: http://www.pniei.penza.ru/RV-conf/T9/S92. (in Russ.)

7. Ivanov A.I., Malygina E.A., Perfilov P.A. et al. The comparison of performance between the criterion mean geometric and the criterion of Cramér-von Mises on a small sample of biometric data. Models, Systems, Networks in Economics, Engineering, Nature and Society. 2016;2:155-158.

8. Ivanov A.I., Perfilov K.A., Malygina E.A. Multivariate statistical analysis of the quality of biometric data on extremely small samples using the criteria of the geometric mean tests calculated for the analyzed probability functions. Measuring. Monitoring. Management. Control. 2016;2(16):58-66. (in Russ.)

9. Ivanov A.I., Vjatchanin S.E., Malygina E.A. et al. Precision statistics: neuron networking of chi-square test and Shapiro-Wilk test in the analysis of small selections of biometric data. Proceedings of the International Symposium Dependability and Quality. 2019;2:131-134. (in Russ.)

10. Ivanov A.I. [Biometric identification of a person based on the dynamics of unconscious movement. Monograph]. Penza: PSU Publishing; 2000. (in Russ.)

11. Ivanov A.I. [Neural network technology of biometric identification of open system users. Author’s summary of the Doctor of Engineering thesis per study program 05.13.01 System analysis, management and processing of information. Penza; 2002. (in Russ.)

12. Malygin A.Yu., Volchikhin V.I., Ivanov A.I. et al. [Fast algorithms of testing neural network mechanisms of biometric cryptographic protection of information]. Penza: Penza State University Publishing; 2006. (in Russ.)

13. Akhmetov B.S., Volchikhin V.I., Ivanov A.I., et al. [Algorithms for testing biometric neural network mechanisms of information protection]. Kazakhstan, Almaty: Satbayev University; 2013. [accessed 14.04.2020]. URL: http://portal.kazntu.kz/files/publicate/2014-01-04-11940.pdf. (in Russ.)

14. Ivanov A.I., Zakharov O.S. [The BioNeiroAvtograf simulation environment: a software product (created by the Laboratory of Biometric and Neural Network Technology, freely available since 2009 on the website of the Penza Electrical Engineering Research and Development Institute)]. [accessed 14.04.2020]. Available at: http://пниэи.рф/activity/science/noc/bioneuroautograph.zip. (in Russ.)

15. Ivanov A.I. [Automatic training of large artificial neural networks in biometric applications: a study guide to laboratory works performed in the BioNeiroAvtograf simulation environment]. Penza: Penza Electrical Engineering Research and Development Institute; 2013. [accessed 14.04.2020]. Available at: http://пниэи.рф/activity/science/noc.htm. (in Russ.)

16. Kobzar A.I. [Applied mathematical statistics. For engineers and researchers]. Moscow: FIZMATLIT; 2006. (in Russ.)

17. Ivanov A.I., Lozhnikov P.S., Bannykh A.G. A simple nomogram for fast computing the code entropy for 256-bit codes that artificial neural networks output. Journal of Physics: Conference Series. 2019;1260(2):022003.