By now, a large number of studies have been conducted in the field of sustainable development forecasting, both in terms of single components and on a country-wide scale, including in accordance with UN regulatory documents. However, there is no consensus on the approaches used, the mathematical apparatus and the sustainability criteria. This research attempts to formulate and formalize the task of forecasting sustainable development, as well as to analyse current experience and select the most appropriate method for forecasting the sustainable development of territories with arbitrary composition and mixed organisational structures. The basic categories and baseline scenarios for sustainable development forecasting are proposed. A forecast accuracy of 10–20 % has been selected as required and sufficient. Linear multiple models, panel data, regression-differential models, including those using the support vector machine, optimisation mathematical models, finite difference models, and machine learning methods are considered. Their advantages and disadvantages are noted. Controversial approaches and concepts from a scientific point of view are critically examined, including the time density of regional development. Requirements for forecasting the sustainability of territory development are also presented, along with strategies and scenarios for their improvement. The study leads to the conclusion that the finite-difference model and artificial neural networks are the most promising of the approaches considered. Although, the drawback of artificial neural networks application is the necessity of experimental selection of the model structure for the task to be solved, the possibility of deadlock situations during training, and the unpredictable nature of the results.
Boyarchuk et al. (Fri,) studied this question.