ELECTRE tree: a machine learning approach to infer ELECTRE Tri-B parameters

Purpose: This paper presents an algorithm that can elicitate (infer) all or any combination of ELECTRE Tri-B parameters. For example, a decision-maker maintain the values for indifference, preference, and veto thresholds, our find criteria weights, reference profiles, lambda cutting level. Our approach is inspired by Machine Learning ensemble technique, Random Forest, that, we named as Tree algorithm. Methodology: First, generate set models, where each model solves random sample alternatives. Each made with replacement, having at least two between 10% to 25% has its parameters optimized genetic use ordered cluster assignment example optimization. Finally, after optimization phase, procedures be performed, first one will merge finding in this way elicitated parameters, second procedure alternative classified (voted) separated model, majority vote decides final class. Findings: We have noted concerning voting procedure, non-linear decision boundaries are generated, they suitable analyzing problems same nature. In contrast, merged generates linear boundaries. Originality: The elicitation technique composed multicriteria models engaged generating robust solutions.