Materials data validation and imputation with an artificial neural network

TL;DR

This paper presents an artificial neural network framework for validating and imputing missing data in material property datasets, effectively identifying errors and leveraging correlations to improve prediction accuracy.

Contribution

It introduces a neural network approach capable of handling incomplete and graphical data, enhancing data quality and validation in materials science.

Findings

Identified 20 errors in a commercial materials database.

Successfully applied the method to alloys and polymers.

Validated the approach with different testing schemes.

Abstract

We apply an artificial neural network to model and verify material properties. The neural network algorithm has a unique capability to handle incomplete data sets in both training and predicting, so it can regard properties as inputs allowing it to exploit both composition-property and property-property correlations to enhance the quality of predictions, and can also handle a graphical data as a single entity. The framework is tested with different validation schemes, and then applied to materials case studies of alloys and polymers. The algorithm found twenty errors in a commercial materials database that were confirmed against primary data sources.