Table 1 Prediction performance benchmarking for the prediction task of ‘Atomistic Line Graph Neural Network (ALIGNN) based Feature Extractor’ on formation energy of JARVIS-3D dataset.

From: Structure-aware graph neural network based deep transfer learning framework for enhanced predictive analytics on diverse materials datasets

Features

Test MAE (eVatom−1)

atom features

0.0465

bond features

0.0625

angle features

0.0754

atom-bond features

0.0410

atom-bond-angle features

0.0449

atom-bond-angle features (last)

0.0401

  1. The table shows the test mean absolute error (MAE) of the best model for each feature type (selected based on validation MAE) when run on features extracted from different layers.
Back to article page