Predictive correlations based on large experimental datasets: Critical constants for pure compounds

Published

July 27, 2010

Author(s)

Andrei F. Kazakov, Chris D. Muzny, Vladimir Diky, Robert D. Chirico, Michael D. Frenkel

Abstract

A framework for development of estimation methods is demonstrated using prediction of critical constants for pure compounds as an example. The dataset of critical temperature Tc and critical pressure pc for over 850 compounds used in the present work was extracted from the TRC SOURCE data archival system and is based exclusively on experimental values taken from the literature. Experimental Tc and pc values were critically evaluated using the methods of robust regression and their uncertainties were assigned in a rigorous manner. The correlations for critical constants were developed based on Quantitative Structure Property Relationships (QSPR) methodology combined with the Support Vector Machines (SVM) regression. The propagation of the experimental uncertainties into the predictions produced by the correlations was also assessed using a procedure based on stochastic sampling. The new method is shown to perform significantly better than a number of commonly used estimation methods.

Citation

Fluid Phase Equilibria

Volume

298

Pub Type

Journals

Download Paper

Local Download

Keywords

correlations, critical parameters, large datasets, QSPR, SVM

Citation

Kazakov, A. , Muzny, C. , Diky, V. , Chirico, R. and Frenkel, M. (2010), Predictive correlations based on large experimental datasets: Critical constants for pure compounds, Fluid Phase Equilibria, [online], https://tsapps.nist.gov/publication/get_pdf.cfm?pub_id=902992 (Accessed October 8, 2025)