An inherent feature of road accident databases is the imbalance between the number of observations associated with accidents with fatal and non-fatal victims of injuries concerning to accidents without victims. This particularity led to the adoption of corresponding balancing techniques, which can resample classes and attributes. Therefore, it ensures that there is no over-adjustment of the data in classification problems. This study investigates the influence of different balancing methods such as undersampling, oversampling and SMOTE on the classification process of road accident severity adopting an Artificial Neural Network approach. The results obtained indicate that all methods used were able to effectively adjust the balance between the minority and majority classes. Balancing leads to a better performance of the classifier, shown by the efficient adjustment of the data to the model, as the gain in the quality and accuracy of the classification process, especially, considering sampling techniques such as SMOTE.
