Accounting for uncertainty of non-linear regression models by divisive data resorting

This paper focuses on building models of stochastic systems with aleatoric uncertainty. The nature of the considered systems is such that the identical inputs can result in different outputs, i.e. the output is a random variable. This paper suggests a novel algorithm of boosted ensemble training of multiple models for obtaining a probability distribution of an individual output as a function of a system input. The deterministic component in the ensemble can be an arbitrarily-chosen regression model. The algorithm does not require any special properties of the model, other than having descriptive capabilities with some expected accuracy for the chosen dataset type. The efficiency of the suggested algorithm is demonstrated by comparing it with the Monte-Carlo simulations and by modelling of a complex social system - sports betting choices adjusted for obtaining a monetary gain.