Outlier Detection, Clustering, and Classification - Methodologically Unified Procedures for Conditional Approach

Piotr Kulczycki, Krystian Franus

Abstract: The subject of the study are three fundamental procedures of contemporary data analysis: outliers detection, clustering and classification. The issue is considered in a conditional approach – introduction of specific (e.g. current) values to the model allows in practice a significantly precise description of the reality under research. The same methodology has been used for all three above tasks, which considerably facilitates the interpretations, potential modifications and practical applications of the material investigated. Using non-parametric methods frees the procedures under investigation from a distribution in the considered dataset.