Gross regional product modeling with machine learning methods

Introduction. Gross regional product is one of the key indicators of regional development; therefore its forecasting modeling is a relevant issue for modern studies. Purpose. The article aims at applying machine learning methods to forecast gross regional product with high accuracy. Materials and Methods. The statistical base included data from periodicals and the Internet, reference materials and annual statistical collections prepared by the public statistics bodies of the Russian Federation. Exogenous variables are 12 factors which reflect the economic development of the region, as well as the level of its digitalization. Machine learning methods are used to model the gross regional product forecast; the Google Colaboratory environment was chosen to write Python codes. Results. This study developed four machine learning models: Linear Regression, Gradient Boosting Regressor, LGBM Regressor, Decision Tree Regressor. Linear Regression has an R2 of 87 % on the test sample, Gradient Boosting Regressor has an R2 of 89 % on the test sample, LGBM Regressor and Decision Tree Regressor have an R2 of 88 % and 81 % on the test sample, respectively, which is a good result. Conclusions. The models described in the paper could be applied to short-term forecasting of gross regional product. The results of the study can be used in socio-economic planning practices, assessing the investment attractiveness of regions, as well as in managing regional policy and developing regional growth policies. Further research may be connected with constructing other machine learning models which could be derived from Lazy Regressor, as well as improving the predictive capacity of the models in this study.