An efficient preference learning method based on ELECTRE TRI model for multi-criteria inventory classification

The multi-criteria ABC analysis is a well known inventory management method for classifying inventory. In the most ABC classification applications, it has been considered fully compensatory approaches, i.e. items have been privilege badly in one or more criteria could be placed in good classes, so it is necessary non-compensatory approach to be noticed. ELECTRE TRI is an outranking relations based model that consider non-compensatory approach, although suffers from the complexity and cost of determining the large number of decision-makers preferences (parameters). In this paper we propose a new method which learns all the decision-maker's preferences from assignment example at the same time using the Particle Swarm Optimization(PSO) algorithm, and will be applied in ABC classification. Against the data mining standard techniques that classify items in nominal way, this model has the ability to categorize items into ordinal classes. The evaluation of proposed method on the illustrated inventory datasets shows high quality and competitive results compared with several standard classification models.