Integrated design of experiments and machine learning for performance prediction and manufacturing condition analysis of complex functional polymer films

Optimizing multi-step polymer film manufacturing involves complex non-linear factor interactions and time-varying process-history variations. To address these challenges under strict proprietary constraints, this work proposes a pragmatic, intelligent modeling framework integrating machine learning (ML) using a small dataset which are designed by Design of Experiments (DoE) approach. After resolving severe multicollinearity via mean-centering, we compared a reduced Response Surface Methodology (RSM) model with Random Forest and Gradient Boosting algorithms. The results show that model performance is highly response-dependent: the interpretable RSM baseline is effective for smoother performance indicators (R2 ~0.65), while Gradient Boosting excels at predicting interaction-sensitive properties (R2 ~0.78). Multi-response target screening successfully mapped a narrow, highly sensitive feasible region governed by competing performance trade-offs and recycling degradation. This integrated approach provides engineers and manufacturers with a low-cost, robust workflow for intelligent process control and resource optimization.