OUTLIERS EMPHASIS ON CLUSTER ANALYSIS The use of squared Euclidean distance and fuzzy clustering to detect outliers in a dataset

Gianluca Rosso (1)

1 SIS - SIS Società Italiana di Statistica

Outlier is the term that indicates in statistics an anomalous observation, aberrant, clearly distant from others collected observations. The outliers are the subject to animated discussions in various contexts with regard to be or not to be considered in the average evaluations. Outliers can become a precious source of information, on condition that be able to accurately identify the presence in the reference datasets. The need to identify the presence of clustered outliers in a dataset not previously treated could argue for a fuzzy clustering, emphasized by using the quadratic Euclidean distance as similarity measure. For interesting and useful results, it should be inclined a possibilistic clustering approach, where the term "possibilistic" means, always in mathematical rigor, a component of interpretation of values that point out anomalous cases. The crisp method does not allow it, the fuzzy method introduce it, the possibilistic one use it. This is a very simple paper with divulgative purposes, addressed especially to students, but not only.

Mots clés: cluster analysis; crisp clustering; fuzzy clustering; squared Euclidean distance; outliers; possibilistic clustering; outliers.

Domaines: Applications [stat.AP]

Gianluca Rosso. OUTLIERS EMPHASIS ON CLUSTER ANALYSIS The use of squared Euclidean distance and fuzzy clustering to detect outliers in a dataset. 2014. ⟨hal-00962248⟩