Investigation of Algorithms for Converting Dimension of Feature Space in Retail Data Analysis Problems

The purpose of this paper is to investigate and determine the most accurate and fast algorithms for reducing the dimension of the feature space in the task of analyzing retail data, because large databases may contain noisy or duplicate information that should be eliminated to improve the quality of data processing. To do this, a review of existing solutions for selection and extract features was conducted, and strengths and old sides were highlighted. The results of this work have shown that the most accurate and fast algorithms are one-dimensional selection, which uses Chi-square as a static criterion, and the method of the scikit-learn SelectFromModel library, which accepts estimates of the parameters of the logistic regression model.