Machine learning to optimize nanocomposite materials for electromagnetic interference shielding

Carbon-based fillers/polymer nanocomposites for electromagnetic interference (EMI) shielding have attracted researchers' attention due to their excellent electrical conductivity and lightweight. The numerous material design features make it flexible to prepare required shielding composites. However, the developments of the composites usually depend on researchers’ experience and repeated experiments, leading to longer development cycles and more costs. This work employs the machine learning approach to establish a shielding effectiveness rapid prediction model and analyze the critical factors and rules in material design, optimizing the new materials development and reducing experiments. Firstly, a dataset of carbon-based conductive particles/polymer nanocomposites for EMI is established, including the most accessible material and structure features. We take Weighted Average Ensemble strategy to ensemble five diverse base models on the dataset and find the final prediction model outperforms all base models. Besides, the importance of features is analyzed by variable importance rankings. The rules that critical features influence the EMI are investigated through model-agnostic techniques: partial dependence and individual conational expectation plots. The prediction model is a valuable tool to predict the shielding performance rapidly. Coupled with rules obtained, the model can guide materials development, shorten the development circus, and reduce costs.