Comparative Analysis of Feature Extraction Algorithms in Investigation of Products Sales Data

The problem of determining the significant characteristics in the observations of the studied objects is considered. A comparative analysis of feature extraction algorithms is carried out, including correlation methods, methods using chi-square criterion, recursive feature exclusion methods and algorithms based on an ensemble of forests of random decision trees (algorithmic descriptions have been formalized according to the subject area). The results of the algorithms - the group of the most important goods characteristics - are additionally analyzed, namely, the selected features are compared: which signs (characteristics of the goods) were chosen by all algorithms and which were not (are there any intersections among the results). With the help of an expert group, we answered the question whether there are any contradictions in the results.