A novel quantitative structure-property relationship model f prediction of lower flammability limits of organic compounds: A combined data splitting-feature selection strategy

A QSPR study is suggested for prediction of lower flammability limits (LFL) of organic compounds. The lower flammability limit (LFL), which is usually in percentage volume (vol.%) at 298 K, is defined as the minimum concentration of a combustible substance that is capable of propagating a flame through a homogeneous mixture of the combustible and the air under the specified conditions of test [1]. Various kinds of molecular descriptors were calculate to represent the molecular structures of compounds, such as chemical quantum, topological, charge, and geometric descriptors [2]. QSPR models are usually obtained by splitting the data into two setsincluding calibration (or training) and prediction (or validation). All model building steps, especially feature selection procedure, is performed using this initial splitting, and therefore the performances of the resulted models are highly dependent on the initial data splitting. To investigate the effects of data splitting on the feature selection in the current article we proposed a combined data splitting feature selection (CDFS) methodology [3] for QSPR model development by producing several different training/validation/test sets, and repeating all of the model building studies.