Bayesian network analysis of accident risk in information-deficient scenarios

José Enrique Martín, Javier Taboada-García, Saki Gerassis, Ángeles Saavedra, Roberto Martínez-Alegría


DOI: 10.7764/RDLC.16.3.439


Analysis of accidents using Bayesian networks links certain predictor factors with other target factors representing types of accidents under study. Databases of real accident reports are typically used for both designing and training networks, which inevitably skews future inferences. Inferences are also limited because such databases do not usually include data on situations where accidents have not occurred. Inferences can thus be made about the occurrence of an accident, but not about specific types of accident. We describe a novel Bayesian network strategy for the field of occupational risk prevention which, extracting data from a database that includes situations where no accident has occurred, quantifies the influence and interactions of factors. It also allows particular accident types to be studied individually, thereby highlighting not only the correlation but also the causal relationship between work setting and accident risk.




Civil engineering, information deficit, Bayesian networks, workplace accident, model reduction.

Full Text:



  • There are currently no refbacks.