Graph neural network for predicting the effective properties of polycrystalline materials: A comprehensive analysis

•Generated a large dataset of over 5000 polygrain microstructures and properties.•Developed a graph neural network (GNN) model with a property prediction error <1.4%•The optimized GNN model outperforms linear regression and two baseline CNN models.•Identified critical and unwanted input features via sequential forward selection.•Demonstrated excellent transfer learning performance of the pretrained GNN model.