Flame retardant properties of metal hydroxide-based polymer composites: A machine learning approach

With increasing the practical applications of polymeric materials, surging demands for the high-performance flame-retardant polymer composites are obvious. Accurately predicting the performance of flame-retardant materials is critical for expediting technology development. Here, we develop machine learning models based on Random Forest algorithm to create standard paradigms based on a self-generated comprehensive dataset consisting of 219 experimental data. Integrated by plenty of randomly combined decision trees, the Random Forest is well known for its accurate predictability and easy interpretability. We predict the important flame-retardancy-evaluating parameters including time to ignition (TTI), peak heat release rate (pHRR), total heat release (THR) and flame retardancy index (FRI) of different polymer composites containing metal hydroxides as the main fire-retardant additive. Our optimized models achieve high accuracy with R-squared value of 0.81 for regression and over 0.85 for classification; mean absolute error is below 0.3 for all 4 models. The analysis of important features reveals that mass fraction of MH takes always the first place, and the influence of feature groups on target properties is usually multi-dimensional and complicated.