Fuzzy-C-Means-Based Classification of Thermodynamic-Property Data: A Critical Assessment

Abstract

We consider the problem of accuracy in the algorithmic classification of thermodynamic properties of fluids from a fuzzy C-means (FCM) clustering methodology. The analysis emphasizes on the way the algorithm is affected by factors such as the natural scale of the data, and the following strategies chosen for the classification: (1) data normalization, (2) transformation, (3) sample size of furnished data, and (4) the value of the fuzzy parameter. Experimental data corresponding to pressure, volume, and temperature of water are taken from the literature and used to show that the natural scaling of the data, the normalization and transformation strategies, as well as the values chosen for the fuzzy parameter are all important factors in the classification. Also, a decrease in the number of data used during the process degrades the quality of the solution. A complete consideration of the issues examined here are undoubtedly beneficial every time a FCM classification is tried on a new problem.

G.A. was the recipient of a PROMEP scholarship from México (PROMEP/PTC-68), for which we are grateful. This project has been partially supported by an NSF-funded CREST center (HRD-0932421), and the CSULA-RSCA program fund.