System Methodology of Data Analysis and Preprocessing for Solving Classification Problems

Abstract

The article describes and investigates the systematic methodology of data analysis and preprocessing for solving classification problems. The methodology combines the following groups of methods on the basis of a systemic approach: methods of processing data gaps, methods of processing anomalous values, methods of feature generation, methods of identifying nonlinearities and non-stationarity, and normalization methods. Methods of character generation were studied. Methods of feature selection and generation are divided into three main groups: filtering methods, wrapping methods, and embedded methods. Since feature selection plays a crucial role in machine learning, increasing model performance and reducing computational costs, the paper proposes a combined feature selection method that includes the step-by-step use of both filtering and wrapping methods. The method consists of five steps to efficiently select the most relevant features of a data set. It offers a better approach to feature selection. This results in improved model performance with fewer features and reduced computational cost. The creation of a red wine classification system was considered for the experimental verification of the system methodology of analysis and pre-processing of data. The Red Wine Quality dataset was used. The purpose of the classification is to identify factors associated with the risk of untimely delivery of previous orders to customers and information about recipients of goods. To solve the task of wine quality classification, simulations were carried out using various algorithms. The effectiveness of the system approach to solving problems of analysis and preprocessing of data to solve problems of classification was proved.