SGC based prediction of the flash point temperature of pure compounds

This work introduces a general quantitative structure property relationship (QSPR) for predicting the Flash Point Temperature (FPT) for 1471 pure compounds. Artificial neural networks (ANN) and multivariable linear regression (MVLR) along with the structural group contribution (SGC) approach were employed to calculate FPT. Several SGC definitions are investigated to predict the desired property based on MVLR. Four structural group contribution methods were proposed based on MVLR resulted in almost the same accuracy with an Average Absolute Error (AAE) ranging from 4 to 5% and a correlation coefficient (R) from 0.93 to 0.96. The ANN method was implemented to enhance the predictions of one of the methods and proved to be the best technique for calculating the FPT of pure compounds. The predicted FPT for the 1471 data set were in good agreement with the experimental values, having AAE of 1.21% and R of 0.9917 using the ANN model. These results were more accurate than other methods in the literature utilizing only the molecular structure of the compounds.