Evaluation of the flammability characteristics of alkyl esters: New QSPR models

Abstract

In the present study, new quantitative structure–property relation (QSPR) models have been developed to predict different flammability characteristics (i.e., Lower flammability temperature (LFLT), Upper flammability temperature (UFLT), Lower flammability percent (LFL(V%), and Upper flammability percent (UFL(V%)) of pure Alkyl esters. In this regard, new data sets containing 179 different alkyl esters from 10 different chemical categories were used. In the model development procedure, the appropriate molecular descriptors were selected for each property using the enhanced replacement method (ERM). Afterward, a multivariable linear model and three nonlinear models based on genetic programming (GP), random forest regression (RFR), and support vector regression (SVR) were developed using the molecular descriptors as input variables. The implementation of different internal and external validation methods confirmed the acceptable prediction capability of the developed models. In this regard, the average absolute error for the best models were 7.61 K for LFLT, 7.96 K for UFLT, 0.10 % for LFL (V%), and 1.10 for UFL(V%) over the whole dataset. A comparison between the prediction capabilities of the developed QSPR models with previous models also confirmed the superiority of the developed models in this study. Therefore, the developed QSPR models can be employed in the evaluation of the flammability characteristics of new alkyl esters such as new fatty acid alkyl esters in the biodiesel.