Methods of cluster analysis in sensor engineering: advantages and faults

• Published in: Semiconductor Physics Quantum Electronics & Optoelectronics
• Date: 2010
• Main Authors: Burlachenko, Yu.V., Snopok, B.A.
• Format: Article
• Language: English
• Published: Інститут фізики напівпровідників імені В.Є. Лашкарьова НАН України 2010
• Online Access: https://nasplib.isofts.kiev.ua/handle/123456789/118565
• Journal Title: Digital Library of Periodicals of National Academy of Sciences of Ukraine
• Cite this: Methods of cluster analysis in sensor engineering:
 advantages and faults / Yu.V. Burlachenko, B.A. Snopok // Semiconductor Physics Quantum Electronics & Optoelectronics. — 2010. — Т. 13, № 4. — С. 393-397. — Бібліогр.: 13 назв. — англ.

We consider the crisp and fuzzy partitioning techniques of cluster analysis
 bearing in mind their application for classification of data obtained with chemical sensor
 arrays. The advantage of the cluster analysis techniques is existence of a parameter S(i).
 This parameter gives quantitative efficiency of classification and can be used as
 optimization criterion for sensor system as a whole as well as the measurement
 procedure. The crisp and fuzzy techniques give practically the same result when
 analyzing the data that cluster uniquely. It is shown that big value of the parameter S(i) is
 not sufficient for adequate data partitioning into cluster in more complicated cases, and
 the results of clusterization for the above techniques may diverge. In this case, one
 should apply both techniques concurrently, checking the correctness of partitioning into
 clusters against the principal component analysis.