Comparative Analysis of Approaches for Solving the Problem of Improving the Quality of Retail Trade Data by Artificial Intelligence

摘要

The purpose of this work is to create requirements for an algorithm that increases the completeness and accuracy of sales data in retail trade. The article analyzes aspects of data quality, as well as the stages at which they are being worked out; a comparative analysis of approaches to eliminating missing values and detecting anomalous values is carried out; criteria for comparison are formulated and existing analogues are described. Based on the results of the comparison, requirements for future algorithms are formed and approaches for solving the problem of improving the quality of sales data are defined. As a result of this work, methods based on simulation were chosen to restore the missing values and methods based on the nearest neighbor algorithm were chosen to detect anomalies. After analyzing the solutions included in the chosen approaches, the algorithms that are most suitable for the given subject area and give the best results compared to the others were selected, namely: for processing missing values – EM-algorithm, for abnormal values – k-NN. Methods for increasing the efficiency of the EM-algorithm and reducing the operating time for the k-NN method are formulated. Directions for further research are suggested.