Study of the Method for Verification of the Hypothesis on Independence of Two-Dimensional Random Quantities Using a Nonparametric Classifier
Aleksandr V. Lapko, Vasiliy A. Lapko, Anna V. Bakhtina · Optoelectronics Instrumentation and Data Processing · 2021
Abstract The properties of the method for verification of the hypothesis on independence of random quantities based on the application of a nonparametric image recognition algorithm corresponding to the maximum likelihood criterion are studied. Estimation of distribution laws in classes is performed using initial statistical data in the assumption of independence and dependence of compared random quantities. In these conditions, estimates for image recognition error probabilities in classes are calculated. Their minimum value is used to make a decision on independence or dependence of random quantities. The proposed method avoids the problem of decomposition of the domain of random quantities into multidimensional intervals. The efficiency of the proposed method with complication of the dependence between random quantities and change of size of initial statistical data is studied via numerical experiments.