Machine learning approach for predicting optical and photothermal properties of gold nanoparticle/polymer hybrid films: Effect of synthetic data

Emerging machine learning (ML) approaches have been adopted in various material systems to predict novel properties with the assistance of the corresponding large datasets. For new materials, however, collecting sufficient data points for model training is not feasible, which is the case for gold nanoparticle/polymer hybrid films. In this study, an ML approach coupled with finite element modeling was proposed for predicting the optical and photothermal properties of gold nanoparticle/polymer hybrid films. Experimental datasets of the optical and photothermal properties were built using results from the literature. Then, finite element analyses were conducted to generate synthetic data to satisfy the quality and quantity of the data required for training models. Correlation analysis and model training were performed using the datasets with and without synthetic data to evaluate their effects on predicting the performance of the ML models. The relative importance of features to targets (properties) was evaluated by correlation analysis. ML models with high accuracy were obtained by training various models from conventional to newly developed algorithms. Advantages, weaknesses, and improvement of the synthetic data addition were discussed. The proposed workflow and framework offer reliable prediction of optical and photothermal properties over different combinations of gold nanoparticles and polymer matrices, which can be extended to include more features related to processing parameters and microstructures.