Quadratic support vector machine learning for modeling and predicting the mechanical performance of sustainable biocomposite materials for structural applications

Abstract

The properties of polypropylene (PP)-based natural fiber materials are influenced by complex nonlinear interactions between chemical composition and structural behavior. This makes the accurate prediction of their performance a challenging task for various industries including civil and structural applications. To address this challenge, this study employs a quadratic support vector machine (SVM) model to predict the tensile strength (TS) and tensile modulus (TM) of PP composites, incorporating a comprehensive dataset of natural fiber reinforcements. The dataset underwent rigorous preprocessing, including outlier removal, feature engineering and data transformation, to enhance the model’s accuracy. The quadratic SVM model exhibited outstanding predictive performance, achieving an R2 of 0.994 for training and 0.990 for testing in TS predictions, and 0.981 for training and 0.98 for testing in TM predictions. The mean squared error values further reinforced the model’s reliability, with 0.00398 (training) and 0.00361 (testing) for TS, and 0.00731 (training) and 0.00723 (testing) for TM. To assess real-world applicability, Simulink-based simulations were conducted, comparing predicted values with experimental measurements. The results demonstrated strong correlation, with minimal deviations observed across various fiber compositions, confirming the robustness of the proposed predictive framework. By leveraging machine learning, this study provides a powerful tool for materials scientists to optimize fiber-reinforced composites, reducing reliance on costly experimental procedures and accelerating the development of sustainable and high-performance materials.