Optimizing a machine-learning model for color design of metal oxides/metal multilayers with physics-guided kernel trick

This study builds and evaluates ML-based prediction models to design the color of metal thin films through oxidation processes. Using images of copper oxides/copper multilayered structures prepared through oxidation of single-crystalline Cu thin films, we apply various regression algorithms, including polynomial regression, random forest regression, and support vector regression (SVR). The best-performing algorithm, SVR with a customized exponential-cosine kernel, highlights the significance of kernel selection based on physics for enhanced performance. Our detailed analyses of challenges encountered in applying ML with experimental data serve as a guide for designing and optimizing the properties of materials using ML.