Iranian Journal of Chemical Engineering (IJChE)

Abstract Polymer nanocomposites reinforced with multi-walled carbon nanotubes (MWCNTs) offer promising mechanical performance; however, predicting their tensile modulus remains challenging due to the complex interplay of multiple factors such as filler content, functionalization, and interphase quality. In this study, a dataset of 229 samples was compiled from the literature, augmented via cubic spline interpolation to 4,933 training points, and analyzed using six machine learning models, including SVR, Random Forest, Gradient Boosting Regressor, XGBoost, KNN, and Artificial Neural Networks (ANNs). The inclusion of the interphase modulus (Ei), calculated via an extended Ji model, proved critical for improving prediction accuracy. Among all models, Gradient Boosting Regressor and XGBoost achieved the best predictive performance (Test R² = 0.9868 and 0.9837, respectively), while ANN demonstrated competitive accuracy (Test R² = 0.9703) but higher sensitivity under cross-validation (Mean CV R² = 0.7486). Feature importance analysis using SHAP further confirmed the significant contribution of Ei to prediction outcomes. Overall, this work demonstrates that incorporating physically-informed features like interphase modulus, combined with robust machine learning pipelines, can substantially enhance the predictive modeling of nanocomposite mechanical properties, providing a valuable tool for material design and optimization.