ВЛИЯНИЕ ЗАШУМЛЕННЫХ ДАННЫХ НА РЕШЕНИЕ ЗАДАЧ КЛАССИФИКАЦИИ ПРИ ИСПОЛЬЗОВАНИИ РАЗЛИЧНЫХ МЕТОДОВ МАШИННОГО ОБУЧЕНИЯ

Abstract

Much attention in scientific publications is paid to the future potential of object classification for various fields of human activity, however, the correct choice of machine learning methods in classification problems with a large amount of noisy data is relevant not only for practical, but also for scientific use. In classification problems, the combination of image analysis and machine learning simulates the process of identifying object(s) and can help the researcher solve effectively any applied problems using machine learning. But in most classification problems there are uncontrolled and random noisy input data, which lead to failures and inaccuracies in solving such problems. Therefore, a classification model developed from a highly specialized dataset is likely to perform acceptably on other similar datasets only if the model can cope with the noisiness of the input data. This article compares the performance of machine learning models: fuzzy association rules, fuzzy decision trees, artificial neural networks, and logistic regression. We examine how these models solve classification problems in the presence of noisy input data. The results of the study showed that fuzzy decision trees models are the most resistant to noisy input data when solving classification problems.