ALGORITHMS OF LEARNING AND EVALUATION OF MACHINE LEARNING MODELS OF STRUCTURED DATA SETS

Abstract

The article deals with the sequential process of preliminary analysis and processing structured data about various types of construction vehicles. The algorithm for building machine learning models, in particular such as linear regression, decision tree and random forest, assessment of the quality of the obtained models and producing results is presented. The work describes research in the field of buying and selling cars on the secondary market using modern data mining technologies. The main objective of this study is to predict vehicle value using attributes highly correlated with price. It is proposed to consider the concepts of pricing by building the following machine learning models: taking into account the characteristics of specific brands of cars, taking into account the characteristics specific to certain types of vehicles, as well as a general model that includes all the characteristics available in the set. Models were built based on linear regression and decision tree methods. The purpose of selecting machine learning algorithms was to minimize errors in cost forecasting, speed of work, and ease of interpretation of the obtained results: based on which data the decision was made and which data have the most significant influence on the formation of the cost. To minimize the prediction error, detailed data analysis and preparation were carried out for each type of construction vehicle. Many experiments were conducted with various methods for finding and removing anomalous observations and for finding and using the essential features. In contrast, such methods as Z-index, interquartile range, recursive removal of features, and feature search based on the detection of dependencies using statistical methods were used. A comparative analysis of the results of each of the models was carried out, and the possible reasons for specific results were analyzed. The problems that arise when solving this regression-type problem are presented – the selection of data that best summarizes the system of formation of the cost of a technical vehicle.