Predictive modeling of fatigue life in natural fiber-reinforced composites using machine learning regression techniques

Natural fiber reinforced composites (NFRCs) have gained recognition as sustainable and environmentally friendly alternatives to synthetic composites. However, estimating their fatigue durability remains a major challenge due to their complex and heterogeneous behaviour. This study develops a machine learning framework to predict the fatigue life of hemp fiber-reinforced HDPE composites from experimental data, addressing the specific limitations of conventional empirical models. The dataset is divided into training and testing subsets following a preprocessing phase that includes logarithmic transformation, normalization, and outlier removal. Four regression models are compared: Multiple Linear Regression, Random Forest, Gradient Boosting Regressor, and Support Vector Regression. Grid Search with 5-fold cross-validation is used to optimize hyperparameters and improve predictive accuracy. Model performance is evaluated using the coefficient of determination (R²), Mean Absolute Percentage Error (MAPE), and Root Mean Square Error (RMSE). Results show that ensemble methods, particularly Random Forest (R² = 99.91%, MAPE = 1.11%) and Gradient Boosting (R² = 99.87%, MAPE = 1.52%), substantially outperform linear models and traditional S-N curve fitting. Feature importance analysis reveals that maximum stress accounts for the majority of prediction variance (approximately 68%–81% depending on the dataset), offering actionable insights for material design. The proposed framework demonstrates strong generalization potential and provides a reproducible template for data-driven fatigue modeling in sustainable composite materials.