Learning and reasoning about road traffic accidents using Bayesian Networks

Kireeva, Irina

Learning and reasoning about road traffic accidents using Bayesian Networks

Citation

Kireeva, I. Learning and reasoning about road traffic accidents using Bayesian Networks. 2022. handle: http://hdl.handle.net/10230/54492

This citation was generated automatically.

Abstract

Road traffic accidents represent a serious problem in Spain due to human losses, medical and productivity costs. This project aims to gain insights about atmospheric and environmental factors that influence accidents, specifically injury severity, with Bayesian Networks. Firstly, a large process of data understanding, cleaning and transformation was performed to choose the suitable data to analyze. Secondly, the correlation between selected variables has been evalu- ated with a non-parametric Chi square test. Thirdly, two Bayesian Networking structures were analyzed with Structure Learning techniques and Kullback-Leibler divergence and compared, where BN-6 is an enriched model of BN-4. Finally, a Variable Elimination algorithm was used to make inferences about the probabilities of injury severity with unknown variables. It was ob- served that visibility and lightning conditions have the greatest influence on the target variable. Nevertheless, the two networking structures did not present any significant differences.