Multiple Criteria Decision Aiding (MCDA) methods are considered a suitable tool for classifying the objects of a decision (such as, for example, actions, alternatives, or options) into different categories (classes or groups). In the context of risk management, techniques that can classify risks and risk factors into different categories can be very relevant for the activities of risk assessment. In this work, we propose a new MCDA method for nominal classification problems. A multiple criteria nominal classification problem exists when the categories are pre‑defined, but no order exists among them. Addressing a problem of that kind consists, therefore, of assigning each action, assessed according to its performances on multiple criteria, to at least one category. The method we propose follows a decision aiding constructive approach. Each category is characterized by several reference actions, assuming that all the used criteria are significant for the decision under consideration. Thus, these reference actions should be previously defined, through a co‑constructive interactive process between the analyst and the decision‑maker (which, in an application to risk management, means a collaboration between a risk expert that is aware of the method, and the “owners” of the domain, to create, depending on the defined risk context, categories for, for example, risks, controls, risk events, consequences, and so on). After that, the assignment of a specific action of the domain of the problem to a category depends on the comparison of such an action to the reference actions according to a membership degree. The membership degrees are defined for each category by the decision‑maker, as expert in the domain. The main properties of the method and their proofs are provided. Some potential applications are introduced in order to show the main features of nominal classification problems. A case study is also presented to illustrate how the proposed method can be applied in the scope of risk management, as also to stress its potential and limitations.
Acknowledgments
This work was supported by national funds through Fundação para a Ciência e a Tecnologia (FCT) with reference UID/CEC/50021/2013. The first author acknowledges financial support from Universidade de Lisboa, Instituto Superior Técnico, and CEG‑IST (Ph.D. Fellowship).