Different techniques, such as: Regression-type generalized linear models, Logit/Probit models, ordered Logit/Probit models have been used to achieve these objectives [2, 3, 4].
However most of these models have their own model assumptions and pre-defined underlying relationships between dependent and independent variables [5]; if these assumptions are violated, the model could lead to erroneous estimations, for example of the likelihood of severity accident.