Premium
Simple Neural Network Models for Prediction of Physical Properties of Organic Compounds
Author(s) -
Prasad Y. J.,
Bhagwat S. S.
Publication year - 2002
Publication title -
chemical engineering and technology
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.403
H-Index - 81
eISSN - 1521-4125
pISSN - 0930-7516
DOI - 10.1002/1521-4125(20021105)25:11<1041::aid-ceat1041>3.0.co;2-5
Subject(s) - boiling point , artificial neural network , unifac , physical property , group contribution method , biological system , volume (thermodynamics) , simple (philosophy) , organic compound , chemistry , thermodynamics , computer science , artificial intelligence , organic chemistry , activity coefficient , physics , phase equilibrium , philosophy , epistemology , biology , aqueous solution , phase (matter)
Quantitative structure–performance‐based neural network models for estimating physical properties of organic compounds are proposed. Several configurations of neural networks with various molecular descriptors as inputs have been tested. The molecular descriptors studied are UNIFAC group surface area and group volume parameters, molecular weight and normal boiling point. The advantage of choosing these parameters is that they can be easily obtained for a given compound, irrespective of the functional groups it may contain. Physical properties estimated include critical temperature, pressure, volume and normal boiling point. Neural network models were found to give reasonable estimates for various classes of organic compounds, often even without recourse to experimentally measured values of any property as an input