Machine learning-guided design of cellulose-based hydrogels for predictive swelling and impact resistance

Robust hydrogel composites were fabricated through a one-step chemical gelation route using carboxymethyl cellulose, acrylamide, and a fixed gold nanoparticle-containing formulation platform to obtain impact-resistant materials with composition-dependent swelling and mechanical responses. Increasing cellulose content progressively reinforced the network, raising tensile strength from 3.68 to 10.79 MPa, increasing compressive strength by 3.3 MPa, and enabling complete resilience recovery of 100%, while dynamic mechanical analysis showed concurrent increases in storage modulus and loss modulus, consistent with improved energy storage and dissipation. Wettability and swelling were likewise governed by network density, with cellulose-rich formulations showing lower equilibrium water uptake and higher water contact angles, whereas swelling remained stimulus-responsive, decreasing at 65 °C relative to 25 °C and increasing markedly at pH 13. Swelling kinetics were well described by Higuchi and Korsmeyer-Peppas behavior, indicating coupled diffusion and network relaxation during water uptake. To predict swelling ratio, a machine learning workflow was implemented using grouped cross-validation that held out complete formulation, pH, and temperature condition groups, with formulation percentage, pH, temperature, and time as inputs. Under this conservative protocol, the random forest model delivered the best mean fold performance, with a mean R² of 0.652 ± 0.257, while the aggregated held-out prediction series reached an R² of 0.842. The CEL₄ composite also protected glass from shattering in bottle, ball, and hammer impact demonstrations, highlighting strong cushioning behavior. These results establish cellulose-tuned p(CEL/AAm)/Au hydrogel composites as promising protective layers with predictable swelling behavior, while Au nanoparticle-specific contributions remain to be isolated by future control studies.