Integrated machine learning approach for final weight prediction in stimuli-responsive four-dimensional printed single-layer elastomer strip

This paper presents an integrated machine learning (ML) framework for predicting the final weight of four-dimensional printed (4DP) single-layer elastomer strips fabricated via material extrusion. Strips composed of Polyurethane (PU), Eudragit® S100, and Polyethylene Glycol 400 (PEG 400) were engineered to respond to pH and temperature through swelling and heat-induced transformations. Experimental analysis showed a strong correlation (r = 0.997) between the initial and final weights, with an initial weight contributing 98.85 % to the predictive accuracy and a thickness of only 0.03 %. Five ML models were evaluated using design attributes such as thickness, pH, swelling time, and initial weight. The Decision Tree (DT) achieved the highest accuracy (>99.95 %), followed by Random Forest (RF) and Extreme Gradient Boosting (XGBOOST) (>99.90 %), outperforming the Artificial Neural Network (ANN) (>98.30 %). The proposed framework highlights the superior performance of ensemble methods in capturing deterministic relationships, offering distinct advantages over physics-based models by combining computational efficiency with a high predictive accuracy.