Improving the Performance of Machine Learning Models by Augmenting Composite Coating Bath Design Data Using a Gaussian Noise-Based Data Augmentation Method

In this study, a novel approach based on data augmentation through the addition of Gaussian noise to limited experimental data was employed to support the inverse optimization of bath parameters for achieving the targeted properties of composite coatings, and the impact of this approach on the performance of machine learning (ML) models was investigated. The study began with the initial limited experimental data obtained in accordance with the Taguchi experimental design. Before proceeding with the data augmentation process, the conformity of the dataset's features to a normal distribution was analyzed and verified. Subsequently, Gaussian noise, considered a natural representation of experimental measurement deviations, was added to the original dataset, significantly expanding its size. Furthermore, Analysis of Variance (ANOVA) and two-way interaction plots were employed to identify the key factors affecting coating properties and their interactions. The augmented dataset, along with the original raw dataset, was then used to train and test various ML models. Model performance was evaluated in terms of predictive accuracy and robustness. The successful predictive performance achieved through data augmentation and ML models provides a foundation for future inverse optimization of bath parameters in line with the targeted coating properties.