Sustainable innovation: Artificial Intelligence-assisted design of bio-based plastics

Abstract

This work presents a machine learning-driven methodology for bioplastic formulation design, valorizing eggshell waste as reinforcement. Two bioplastic formulations were developed: bioplastic from Droguería Industrial San Juan starch (BADSJ) and bioplastic from Biopack starch (BAS), reinforced with eggshell powder (ESP). PCA was performed on thickness, moisture, solubility, puncture strength (Ps), deformation at break point (Pd), biode-gradability and opacity. Six machine learning algorithms were evaluated. Random Forest ntree = 200 and Linear Support Vector Machine (SVM) achieved the best performance, achieving 83.3% accuracy and 0.81 Kappa. These models estimated concentration ranges associated with improved Ps and Pd. The SVM exhibited higher pre-dictive consistency, with R2 of 0.9724 and lower error metrics (RMSE: 0.1422; MAE: 0.1043) compared to Random Forest. SVM predicted broader range of eggshell concentrations (0.47–1.47%) associated with improved Ps and Pd values, indicating better differentiation between formulations. These models provide a data-driven framework for optimizing bioplastic formulations and reducing experimental iterations.