Assessment of the impact of education in Russian conditions using machine learning methods

Subject. The article explores possibilities of using machine learning methods to assess the impact of education. Objectives. The purpose is to assess the impact of the level and profile of education using data of the Russia Longitudinal Monitoring Survey – Higher School of Economics (RLMS-HSE) and machine learning methods. Methods. Linear and logistic regression, k-nearest neighbors algorythm, support vectors, decision trees, random forest, boosting, and fully connected neural networks are used to predict wages, job satisfaction, and unemployment risk. Gender, age, region of residence, level and profile of education are used as explanatory variables. Results. The peak of salaries in the Russian labor market is achieved at a relatively young age. The highest salary premium for men is provided by higher education in the field of medicine and information technology, for women – in the field of medicine. The most job satisfaction among men with higher education is experienced by IT and medical graduates, among women – by graduates of pedagogical and agricultural profiles. Conclusions. The use of machine learning methods did not significantly improve the estimates of the impact of the level and profile of education obtained under the J. Mintzer model, although the boosting models showed a lower standard deviation in the test sample compared with the forecasts of regression models. When building financial models focused on comparing the cost of education with the increase in income from work, machine learning models are preferable due to higher accuracy of forecasts.