Destructive tests are typically used to evaluate the fire performance of polymers
and their composites, implying high material costs and long testing times. Developing
numerical models to predict flammability requires advanced mathematical expertise,
IT resources, and realistic input parameters. In this study, we aimed to predict the
key flammability parameters based on the chemical structure of the resin matrices
and fibre content of composites, providing a potential alternative to costly experimental
methods. We employed Random Forest Classifier (RFC), XGBoost algorithms, and an artificial
neural network (ANN) model to predict key combustion parameters: peak heat release
rate (pHRR), time to ignition (TTI), total heat release (THR) and the char residue
(CR) solely based on chemical structure of the epoxy matrix and fibre content of the
composite. After making the predictions, we assessed the performance of the models
using consistent statistical indicators (mean absolute error (MAE), mean square error
(MSE), and the determination parameter (R2)).
