ESTIMATION OF INCOME OF AN ENTERPRISE

Abstract

The article considers the task of increasing the reliability of estimating the level of income of a metallurgical enterprise. The dependence of the enterprise's income on the following factors is analyzed: average annual value of non-current assets; number of employees; cost of goods sold. Using the least squares method, linear and nonlinear univariate and multivariate models were constructed. Due to the presence of multicollinearity between factors, univariate models using the least squares method were constructed for each of the factors. And multivariate models were constructed for the following factors: average annual value of non-current assets; number of employees. Multicollinearity between the specified factors is within acceptable limits. To test the internal parameters of regression models constructed using the least squares method for significance, the magnitude of their p-values was used. It was found that only three single-factor models can be used to predict the net income of a metallurgical enterprise. These are the logarithmic, polynomial, and power model based on the factor - the cost of goods sold. In multivariate models, using the least squares method, the average approximation error exceeds 10%, which indicates the impossibility of their use. At the next stage, it was decided to apply the methods of machine learning with a teacher: k-nearest neighbors and random forest. With their help, single-factor and multifactor models were built. Due to the high average approximation error, all models using the k-nearest neighbors method turned out to be unacceptable. According to the results of comparing the characteristics of the obtained models, the best model was identified. This is a single-factor model, which is built using the random forest method for the factor - the cost of goods sold. The average approximation error of this model was 6.5%. Good values of this indicator were also found in the multifactor random forest model, where its value was 7.4%. As a result of the research, it is recommended to use a single-factor model based on the cost of goods sold factor, which is built using the random forest method, to predict the income of a metallurgical enterprise.