Machine learning-assisted formulation design of flame-retardant polyoxymethylene composites
Year: 2026
Type: article
Abstract: To accelerate the design of flame-retardant formulations and mitigate the time and cost constraints inherent in conventional trial-and-error approaches, this study employs machine learning to facilitate the formulation design of flame-retardant polyoxymethylene (POM) composites. A predictive model for flame-retardant performance was developed using the limiting oxygen index (LOI) as the target variable. A dataset comprising 221 POM-based flame-retardant composite formulations was compiled from peer-reviewed literature, open-source databases, and experimental data generated by our research group. Prior to model development, feature engineering and feature selection were conducted. Six machine learning algorithms—namely, linear regression, Lasso regression, Ridge regression, k-nearest neighbors, random forest, and extremely randomized trees—were evaluated and compared. Among these, the random forest model exhibited the highest predictive accuracy, achieving initial R 2 values of 0.8833 and 0.8744 on the training and test sets, respectively. Following hyperparameter optimization via grid search, model performance improved further, with R 2 values increasing to 0.8938 and 0.8879, accompanied by a reduction in prediction error. Experimental validation using the D-A-BOZ and J-B-BOZ flame-retardant systems confirmed the reliability of the optimized model in predicting formulations within the dataset; however, its predictive capability for novel components remains constrained. Overall, this work demonstrates the potential of machine learning to elucidate the relationship between formulation composition and flame-retardant performance, thereby offering a data-driven strategy for the efficient design of POM-based flame-retardant composites.
Source: Review of Materials Research
Institutions: Henan University, Xiamen University
Language: English
Subfield: Polymers and Plastics
Field: Materials Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $1,600
OpenAlex ID: W7204254020