Creation of a strategy for the innovative development of industry in the regions of Russia using machine learning

Machine learning technology is a powerful tool for analyzing big data, and thus they can be applied to create a strategy for innovative development of industry in the regions of Russia. To compare the results of the quality of forecasting and to choose the most optimal method on the example of radio-electronic industry (REI), the authors applied the method of “k-nearest neighbor machine learning†, neural networks “multilayer perceptron†and adaptive neuro-fuzzy inference system which includes a particle swarm algorithm as well as multidimensional adaptive regression splines. The following functions were studied as the target ones: 1) the volume of innovative goods; 2) developed advanced production technologies; 3) net ï¬ nancial result (informatization and communication). The suggested model was trained on the basis of a sample package of nine inputs and three targets in 2010-2022 in 83 regions of Russia. For further veriï¬ cation of the trained model the year of 2023 was chosen as a sample package. It was stated that the highest quality forecast was made with the k-nearest neighbors machine learning algorithm. During the assessment, it was established that the prospects for innovative development in the industry can be found in the regions where the values of the predicted target functions fall into the planned segments in 2023. The assessment was conducted only in those regions whose forecast was regarded as the highest quality (with the average absolute percentage error < 0,5). These regions include the Krasnodar and the Perm territories, and Nizhny Novgorod, Sverdlovsk, Chelyabinsk and Novosibirsk regions. When performing similar analysis for algorithms of multidimensional adaptive regression splines (for target 3), particle swarm (for target 2), multilayer perceptrons (for target 1), the authors established that Nizhny Novgorod and Sverdlovsk region can be regarded as leaders in the REI industry, and this partially conï¬ rms the conclusions obtained by the machine learning method.