Biometric stats: smoothing histograms based on small training sample

The question of improving the stability of statistical calculations with small training set by complications statistical processing raw data. It is shown that the accuracy of the estimation of the distribution law of small samples of biometric data can be increased by smoothing the histogram. It is proposed to use some digital filter, which will be smoothed traditional histograms, and due to this additional processing will improve the stability of the statistical calculations. The correct choice of the window digital anti-aliasing filter and multiple artificial increase in the number of discrete, used in digital filtering allows to significantly increase the power of goodness of fit test Gini and chi-square test. Graduated Gini goodness of fit test Gini is less sensitive to the number of examples in the test sample. Thus, its use leads to a decrease in error probability of the second kind, due to the limited number of experiments. The article presents a comparative table of the error probability of decision-making by the chi-square test and the Gini for the smoothed data. In contrast to the chi-square test, the Gini criterion is functional even on a sample of small kolichistvo experiments. Thus, for the problems we are seeing biometrics obvious benefit from the application of the criterion of Gini.