Modelling of halomethanes using neural networks

Boiling points at normal pressure of forty-eight halomethanes were investigated. Three modelling methods - partial least squares (PLS), quadratic PLS (QPLS) and neural networks (NN) - were used to evaluate the performance. First a linear modelling method, PLS, was employed to investigate the behaviour of the boiling points. The linear modelling method was not sufficient because the boiling points of the halomethanes had a non-linearity. In order to take its non-linearity into consideration, QPLS was employed, resulting in significant improvement. Finally NN was employed and then the best results were obtained of the three. NN showed a capability which was accurate enough to predict the boiling points of unknown compounds.