Prediction quality of Bayesian belief network model for risky behavior: comparison of subsamples with different rates

Abstract

The study investigates the proposed approach for behavior modeling on the base of Bayesian belief networks that allows predicting behavior characteristics using small and incomplete data from surveys about behavior episodes. We explored the prediction quality of the models in case of rare behavior. The test dataset was automatically generated and included 24465 cases. During the experiment, we considered cases with different rates to compare prediction quality. Our findings suggest that the model had a good prediction quality especially for rare and frequent behaviors (about 92% accuracy) and lower measures for medium-rate behaviors (about 86% accuracy).