Flat superhydrophilic coating and machine learning of co-sputtered coatings for wettability prediction

Co-sputtering is a widely adopted technique for producing flat coatings with tailored wettability; however, it typically requires time-consuming and iterative tuning of deposition parameters through trial-and-error experiments. In this study, machine learning (ML) models are employed to predict the water contact angle (WCA) of co-sputtered coatings. A dataset comprising 86 samples was used to train and evaluate four ML models; of these, 73 data points were extracted from the literature, and 13 were experimentally created by depositing thin films on silicon wafers through co-sputtering. During the experimental data generation, a novel flat coating composed of silicon (Si) and magnesium (Mg) was developed, exhibiting superhydrophilic behavior with a WCA below 5°. The coating demonstrated short- and long-term durability, retaining its superhydrophilic characteristics for up to 30 days under ambient laboratory conditions. Remarkably, even after one year, the surface could recover its superhydrophilic state following basic chemical cleaning. To enhance model performance, the hyperparameters of each ML model—linear regression, decision tree, random forest, and gradient boosting regression—were systematically optimized. Hold-out validation as well as 10-fold cross validation approaches are used to investigate the performance of each ML model. Among the models tested, random forest exhibited the highest prediction accuracy.