Artificial neural network prediction of glass transition temperature of fluorine-containing polybenzoxazoles

Abstract

Fluorine-containing polymers belong to high-performance polymers with unique chemical and physical properties that are not observed with other organic polymers. In this article, three structural parameters were used to correlate with glass transition temperature T_g values for 52 fluorine-containing polybenzoxazoles. The descriptors obtained directly from the structures of repeating units can reflect the chain stiffness (or mobility). Back propagation artificial neural network (ANN) and multiple linear regression (MLR) analysis were used in the study. The final optimum neural network with [3-1-1] structure produced a training set root mean square (rms) error of 2.35 K (R = 0.980) and a test set rms error of 2.30 K (R = 0.978). The statistical results indicate that the ANN model given here has better predictive capability than other existing models.