ABSTRACT
A global dataset of refinery accidents for the years 1990–2016 was analyzed to evaluate the capacity of 16 attributes to differentiate between accidents that cause or not fatalities. For this purpose a Dominance-based Rough Set Approach (DRSA) analysis was carried out. The quality of approximation and accuracy measures confirmed that the established information table is able to distinguish outcome levels in terms of fatalities. Furthermore, the suitability of the extracted rules to describe hidden relationships in the accident dataset was demonstrated. Although, the predictive capacity of the decision rules was not satisfactory, the rules still proved to be useful to identify the attributes that contribute most to assign an accident to the correct outcome class. In summary, this study provided a number of new and substantial insights on worldwide refinery accidents, which complement and extend previous findings for accident frequencies and associated trends as well as different types of consequences.
