Data-driven prediction of flexural and impact behavior of polymer composites

Abstract

With the growing demand for advanced polymer composites, intelligent prediction methods offer an efficient route to optimize material performance while minimizing experimental effort and cost. This study focuses on fly ash and Azadirachta indica reinforced polymer composites, where regression-based machine learning models were employed to predict key mechanical properties, specifically flexural and impact behaviors. Several models, including LASSO, Ridge, Elastic Net, Bayesian Ridge with Automatic Relevance Determination, and Huber regression, were developed and evaluated. Without feature engineering, LASSO and ARD demonstrated strong predictive performance, while Ridge and Huber showed weaker generalization. Incorporating feature engineering significantly improved prediction accuracy and robustness, particularly for flexural and impact behavior. The proposed ML framework demonstrates the potential of intelligent prediction in guiding composite formulation, thereby reducing experimental trials and accelerating the development of high-performance polymer composites.