Identification And Detection of Outliers in Multicriteria Decision Aid.

Abstract

Outlier detection, also known as anomaly detection, is a crucial process in data analysis used to identify data points that significantly deviate from the general pattern of a dataset. The problem of outlier detection in the Multicriteria Decision Aid (MCDA) field has not been extensively explored in the current literature. As far as we know, only three papers have addressed this topic. This study proposes a novel approach for outlier detection in MCDA. The contribution is twofold. The first part introduces the importance degree, a measure that captures the strength of a preference relation between alternatives in PROMETHEE. Each alternative is represented as a vector of its preference values with respect to all other alternatives. The importance degree is calculated using the Euclidean distance between these vectors, providing a clear metric for how strongly one alternative dominates another. The second part of the contribution applies this concept to outlier detection in MCDA. The proposed approach evaluates how similar an alternative is to others by aggregating importance degrees, thus identifying how typical or atypical it is in the decision space. Alternatives with significantly low similarity are flagged as outliers using statistical tools such as the interquartile range or the standard deviation method. Together, these contributions offer a novel framework for both modeling preference intensity and improving the detection of anomalous alternatives in MCDA. This enhances the reliability and interpretability of decision support systems that rely on outranking methods like PROMETHEE