Reconstructing Unknown Coefficients of Stochastic Differential Equations and Intelligently Predicting Random Processes with Directed Learning

A way of intelligently predicting random processes is described, based on more complete use of information about statistical patterns of the evolution of an observed process. At the stage of training the predictive algorithm, the feature space is enriched with the parameters of mixed probabilistic models that allow the construction and reconstruction of the coefficients of a stochastic differential equation describing the given random process. The use of additional statistical information imposes additional conditions on the search area and therefore narrows the set of considered options. Learning is thus targeted by excluding impossible or unlikely options in advance, making it more effective and forecasts more accurate.