Solving local convergence problem and missing value problem in clustering by validated numerics

2014 Fiscal Year Final Research Report

• Project/Area Number: 24500282
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Research Field: Sensitivity informatics/Soft computing
• Research Institution: Shibaura Institute of Technology
• Principal Investigator: KANZAWA Yuchi 芝浦工業大学, 工学部, 准教授 (00298176)
• Project Period (FY): 2012-04-01 – 2015-03-31
• Keywords: ファジィクラスタリング

Outline of Final Research Achievements

The purpose of this research project is to introduce validated numerics techniques to clustering, to solve the local convergence problem in clustering and to deal with the uncertainty of given data for clustering.
In local convergence problem, the all solution algorithm in validated numerics could not produce its result for entropy-regularized fuzzy c-means (eFCM) within adequate time because eFCM has many local optimal solutions. On the other hand, it was clarified that a maximizing model of Bezdek-type fuzzified c-means algorithm can be transformed into a trace maximizing problem naturally, which can be solved globally.
In dealing with the uncertainty of give data, an algorithm was constructed to represent given missing values as the interval with infinite width, and to execute clustering along with divide such the intervals with the aid of all solution algorithm in validated numerics.

Free Research Field

クラスタリング