Forecasting based on Bayesian type models

Abstract

A review of some Bayesian data analysis models is proposed, namely the models with one and several parameters. A methodology is developed for probabilistic models construction in the form of Bayesian networks using statistical data and expert estimates. The methodology provides a possibility for constructing high adequacy probabilistic models for solving the problems of classification and forecasting. An integrated dynamic network model is proposed that is based on combination of probabilistic and regression approaches; the model is distinguished with a possibility for multistep forecasts estimation. The forecast estimates computed with the dynamic model are compared with the results achieved with logistic regression combined with multiple regression. The best results were achieved in this case with the combined dynamic net model.