Clustering of Russian Manufacturing Companies by Indicators of Their Financial Condition Using Machine Learning Technologies

Clustering of research objects and combining them into similar groups based on a set of characteristics is an important stage in solving many tasks of socio-economic development, especially tasks related to assessing the state of the socio-economic system, as well as modeling and forecasting indicators of its future development. The purpose of this study is a relative assessment of the financial condition of large Russian manufacturing companies based on data from accounting reporting forms using clusterization methods classified as machine learning without a teacher. The results of such an assessment are subsequently supposed to be used to build a model for assessing the financial condition of companies based on one of the machine learning algorithms with a teacher. The paper offers key indicators of the financial condition of companies, on the basis of which it is proposed to perform their clusterization. They were identified as a result of the analysis of modern methods and approaches to research and assessment of competitiveness and competitive position of companies. When conducting clusterization based on the proposed set of indicators, financial reporting data from 2,249 Russian manufacturing companies based on the results of 2023 were used. Companies with a turnover of more than 2 billion rubles and a staff of more than 251 people were considered as large companies.. K-Means++, hierarchical clustering, and DBSCAN were used as clustering algorithms. In order to obtain the best result, special data preprocessing and selection of the necessary hyperparameters for clustering algorithms were carried out. The quality of the final clustering was assessed using the Davies-Bouldin (DBI) and the Calinski–Harabasz (CHI) scores. The results showed that the production companies under consideration can be combined into a relatively small number of clusters (usually no more than 3) in terms of financial condition, which opens up wide opportunities for building models of the financial condition of companies. Based on the results of using 3 clustering methods, K-Means++ turned out to be the best algorithm by a small margin, the formed centroids of which can be called the average assessment of companies with poor, normal and good financial condition. The quality of the final clustering can be assessed as good.