Machine Learning-Driven prognosis of tensile strength in FDM printed PLA at various raster angles

Additive manufacturing by fused deposition modelling enables rapid and sustainable fabrication but suffers from anisotropic mechanical performance, particularly in polylactic acid. In this study, the ultimate tensile strength of polylactic acid samples printed using fused deposition modelling was systematically investigated across 108 specimens fabricated from 27 raster angle sequences under four build orientations. Tensile testing revealed ultimate tensile strength spanning from as low as 3.89 MPa in Z-axis builds to 42.67 MPa in X-axis aligned raster angles, underscoring the critical role of deposition geometry. Scanning electron microscopy confirmed fracture modes transitioning from brittle interlayer delamination at low strengths to cohesive bead rupture at high strengths. To overcome the inefficiencies of trial-and-error parameter optimization, three supervised regressors, namely, Random Forest Regression, Support Vector Regression, and Extreme Gradient Boosting (XGBoost), were tested within leak free pipelines using engineered descriptors including raster angle, build orientation, quadratic, and interaction terms. Comparative evaluation with rank-sum aggregation across R², RMSE, and MAPE identified XGBoost as the best fit model (R² = 0.9712, RMSE = 2.14 MPa, MAPE = 8.97%), exhibiting strong calibration and minimal generalization gap. This integration of controlled experimentation with explainable machine learning establishes a predictive methodology that reduces design iteration, strengthens process understanding, and provides a transferable methodology for forecast driven additive manufacturing of polymers.