Abstract
A quantitative structure-property relation model was developed to predict the upper flammability limit (volume percent in air) of pure compounds, based only in their structures. A group contribution method was used to determine the upper flammability limit (UFL) through an artificial neural network. The method was used to identify the structure groups that have a significant contribution to the target property and concluded that 30 atom-type structure groups can represent the UFL for 550 pure substances. The models input parameters are the number of occurrence of each of the 30 structure groups in the molecule. The model predicts UFL with a correlation coefficient of 0.9996 and average absolute deviation of 0.17 vol %. The results were compared with multivariable regression model and other methods in the literature. The model is very useful and convenient for accessing the hazardous risk potential of chemicals for which experimental data is not available.