Toward a supervised machine learning approach to predict the properties of lignocellulosic filler/PHBV-based biocomposites

Biocomposites made from polyhydroxyalkanoates and lignocellulosic fillers are gaining traction as biodegradable and sustainable materials that can be used in applications such as vine clip or food packaging. In this study, supervised machine learning was used to predict the performance of several polymer-based biocomposites. An original dataset compiled from experimental trials was analyzed using classification models to predict the degradation of six main functional properties (mechanical properties, water vapor permeability, and thermal stability) related to filler content and key intrinsic properties (biochemical composition and apparent median diameter (D 50 )). The best-performing machine learning models were evaluated using the F1 score to ensure accuracy, precision, and model reliability. SHapley Additive exPlanation (SHAP) values were used to enhance interpretability and revealed the dominant role of filler content and size on most of the composite properties. As a complement to these results, an abacus was also created to provide an intuitive visualization that could suggest new experimental trials based on specific requirements for the final biocomposites. This integrative, data-driven approach not only advances the predictive design of eco-friendly biocomposites but also offers a novel framework for optimizing the performance of materials while minimizing environmental and production costs.