Forecasting models using artificial intelligence in sectoral economy

The instruments of economic regulation have a significant impact in managing the social standard of living in the country and in the global community. One of the priority areas of economic management is the sectoral economy. To date, there are quite a large number of forecasting methods in the economy and, in particular, in the sectoral economy. The methods vary according to the forecasting objectives and input data. Statistical econometric methods and artificial intelligence methods are considered the most popular. This article is devoted to forecasting in the agricultural sector of the economy using artificial intelligence methods. Several approaches are considered, consisting of a predictive block and a combination of empirical and predictive blocks. The system of crop simulation AGROTOOL is used as an empirical block. The predictive block uses three machine learning methods: the Random Forest algorithm, Ridge Regression method, and Lasso Regression method. These machine learning methods were chosen due to the multicollinearity of the data. Thus, the random forest algorithm was considered in two variations: with and without the Principal Component Analysis (PCA) method. In this paper, in addition to analyzing predictive models, the author also analyzed the selection of key parameters of the random forest PCA and ridge regression: n_components and α methods, respectively. Based on the results of numerical experiments, it can be concluded that depending on the input of the forecast block of data, which may be the values of natural factors (soil moisture content, temperature, solar radiation, average daily specific humidity, specific humidity of leaf saturation, etc.), or the results of formula calculations, the most effective methods of machine learning are random forest and ridge regression. The use of the ridge regression method is most effective with the data preprocessed by AGROTOOL. At the same time, the random forest method produced a small fractional error when forecasting without AGROTOOL and with the combined data.