Prediction of mechanical properties of carbon fiber/epoxy composite modified by nanoparticles using multiple explainable machine learning algorithms

Prediction of Mechanical Properties of Epoxy Composite modified by Nanoparticles Machine Learning. • ML prediction of mechanical properties of nanomodified CF/epoxy composites. • Various machine learning models were used to predict the most accurate tensile and flexural strength. • SHAP-based explainable AI analysis revealed the key influence factors on mechanical strength prediction. • The proposed framework improves predictive accuracy, enabling data-driven optimization. The complex interdependencies between constituent materials and processing conditions of Carbon fiber-reinforced epoxy composites enhanced with nanoparticles hinder the accurate prediction of their performance. This study introduces an interpretable machine learning framework to predict tensile and flexural strength in nanomodified CF/epoxy composites. A comprehensive dataset comprising 174 tensile and 108 flexural data points was curated from 50 peer-reviewed studies, encompassing 12 key input parameters such as fiber layers, nanoparticle size and type, and processing conditions. Four regression models were trained using 5-fold cross-validation and assessed via R 2 , MAE, and RMSE. The Decision Tree and Gradient Boosting models achieved the highest test R 2 scores of 0.983 and 0.931 for tensile and flexural strength, respectively, while XGBoost showed consistent performance across both. To overcome the black-box nature of these models, SHapley Additive exPlanations (SHAP) were employed to elucidate feature importance and interactions. For tensile strength, fiber layer count, sonication temperature, and curing temperature were dominant contributors, while flexural strength was most influenced by curing duration, sonication duration, and particle type. This integrative approach not only enhances predictive accuracy but also delivers actionable insights into process-property relationships, paving the way for data-driven composite design and accelerated materials development in advanced engineering applications.