Abstract

Additive manufacturing (AM) of Acrylonitrile Butadiene Styrene (ABS) components enables the fabrication of lightweight, customizable structures; however, accurately predicting their bending behavior remains challenging due to complex process–property interactions. In this study, key 3D printing parameters were experimentally varied to examine their influence on mechanical performance, while machine learning techniques were applied to model critical outputs, including flexural strength, modulus, hardness, surface roughness, stiffness, and printing time. Several regression algorithms such as Stacking Regressor, Gradient Boosting, Random Forest, and K-Nearest Neighbors (KNN) were evaluated based on R², MAE, and MSE metrics. The tuned Stacking model exhibited the best overall predictive performance, achieving high accuracy for flexural strength (R² = 0.8667) and surface roughness (R² = 0.7641), whereas Gradient Boosting performed best for stiffness prediction and Random Forest for printing time (R² = 0.9974). Flexural modulus and stiffness demonstrated higher prediction errors, suggesting greater modeling complexity. These results underscore the effectiveness of ensemble learning, particularly tuned stacking, in accurately forecasting the mechanical behavior of 3D-printed polymer components, thereby supporting more reliable process optimization and material design in additive manufacturing. 

Keywords 

3D Printing, Additive Manufacturing, Acrylonitrile Butadiene Styrene (ABS), Machine Learning, Ensemble Technique.