Prediction of absolute entropy of ideal gas at 298 K of pure chemicals through GAMLR and FFNN

Thermodynamical optimization for energy conversion system can be performed by decreasing entropy generation. For calculation of entropy, we need to know entropy of ideal gases at 298 K as a reference point. Entropy is a thermodynamic quantity which is not easily measured and prediction of entropy by molecular structures for new designed molecules may be a challenge. An easy and accurate equation for prediction of absolute entropy of pure ideal gas at 298 K was introduced for the first time based on the quantitative structure property relationship (QSPR) approach. Thousand seven hundred pure chemical compounds and 3224 molecular descriptors were used for finding this easy equation by genetic algorithm multi-linear regression (GAMLR) subset variable selection. Our work are based on 1700 chemicals in 81 chemical families that is the most comprehensive available data sets for absolute entropy of ideal gases. The final model is linear and has three molecular descriptors with the squared correlation coefficient of 0.9885 (R2 = 0.9885). Also, feed forward neural network (FFNN) was used for considering non linearity effect of the model. It has the squared correlation coefficient of 0.9909 (R2 = 0.9909). The model passes all validity check methods. The novel proposed model has the predictability for new designed molecules by having the molecular structures of them.