Sustainable Biocomposites Reinforced with Waste Artichoke Stem and Modified Soybean Oil: Multifunctional Properties and ANN-Based Prediction.

Sustainable polyurethane-based biocomposites (PUBs) reinforced with waste artichoke stem (WAS) and modified soybean oil (MSO) provide a promising approach for agricultural-waste valorization and the development of multifunctional polymeric materials. In this study, 48 PUB formulations are prepared by systematically varying WAS content from 0.0 to 3.5 wt.% and MSO content from 0 to 5 wt.%. The dielectric constant, thermal conductivity, Shore A hardness, bulk density, tensile strength, and elongation at break are experimentally evaluated. Across the investigated formulations, the dielectric constant, thermal conductivity, Shore A hardness, bulk density, tensile strength, and elongation at break range from 1.10 to 1.56, 0.024 to 0.036 W m-1 K-1, 9.5 to 43.0, 34.0 to 67.0 kg m-3, 115 to 297 kPa, and 62 to 176%, respectively. A multi-output artificial neural network (ANN) model is developed in MATLAB using WAS and MSO contents as input variables and the six experimentally determined properties as simultaneous outputs. The ANN architecture consists of two input neurons, one hidden layer with ten neurons, and six output neurons. Levenberg-Marquardt (LM), Bayesian Regularization (BR), and Scaled Conjugate Gradient (SCG) algorithms are comparatively evaluated using 42 samples for model development and six independent samples for external validation. The results demonstrate that the ANN successfully captures the nonlinear relationships between formulation variables and the investigated multifunctional properties. Among the evaluated training algorithms, BR provides the most accurate and robust predictive performance, followed by LM, whereas SCG exhibits comparatively lower prediction accuracy. The proposed experimental-computational framework enables reliable simultaneous prediction of the electrical, thermal, physical, and mechanical properties of PUBs and provides an efficient strategy for reducing experimental effort and accelerating the formulation and performance assessment of sustainable biocomposites.