LA-COMET: Toward Reducing Redundant Criteria in Multi-criteria Decision Analysis

Abstract

Multi-Criteria Decision Analysis (MCDA) is an effective tool for decision-making in complex situations characterized by multiple, often conflicting, criteria. However, identifying and determining the relevance of individual criteria can be complicated, especially when the number of criteria is excessively high or when decision experts are not readily available. This paper presents LA-COMET, an innovative method for reducing redundant criteria in MCDA. LA-COMET combines the Characteristic Objects Method (COMET), based on expert preference modeling, with the Regression Lasso technique for automatically selecting relevant variables. Through this combination, our method effectively reduces the number of criteria while maintaining relevant decision-making information. In this method, the Lasso approach also plays the role of an artificial expert, which relies on previous sample evaluations to support COMET model identification. We also present the results of our study, in which we compare the effectiveness of LA-COMET with traditional MCDA methods. In a practical case, the problem of evaluating countries regarding their military potential is considered, where we attempt to re-identify such a model. Our experiments show that LA-COMET achieves significantly better results than classical methods, ensuring high accuracy and relevance of the decisions made.