A1279
Title: Latent feature extraction based on asymmetric degree of reliability for fuzzy clustering
Authors: Mika Sato-Ilic - University of Tsukuba (Japan) [presenting]
Peter Ilic - University of Aizu (Japan)
Abstract: In recent years, the analysis of large-scale and complex data has become increasingly important. A key challenge is how to extract essential latent features from such data and how to derive evaluation values that allow meaningful interpretation. Fuzzy clustering is known as an effective and robust classification method because it incorporates the inherent uncertainty of data classification. However, the interpretation of fuzzy clustering results has often relied on subjective judgment by analysts, and how to perform this interpretation objectively remains an important issue. This problem is also significant in AI methods based on fuzzy clustering and, more broadly, in machine learning approaches. A framework is proposed that defines the reliability of fuzzy clustering results so as to preserve classification uncertainty, using an asymmetric aggregation operator based on both clustering results and individual classification information. This method satisfies norm conditions in a statistical metric space and enables evaluations that reflect the order of human judgment. Using this framework, learning effects associated with the selection of learning tools are extracted and evaluation values that explicitly consider judgmental ordering are presented.
