Machine Learning Approach to Community Detection in a High-Entropy Alloy Interaction Network.

Area of Science:

• Materials Science
• Computational Science
• Data Mining

Background:

• Interaction network analysis is increasingly used across scientific disciplines, including social sciences, health informatics, and biological sciences.
• High-entropy alloys (HEAs) present complex datasets that can benefit from network-based analytical approaches.
• Discovering functional communities within HEA data can accelerate the identification of novel alloy compositions.

Purpose of the Study:

• To construct interaction networks using high-entropy alloy (HEA) descriptors to identify functionally similar HEA communities.
• To explore the potential of these communities for predicting new alloys without experimental testing.
• To compare the effectiveness of the Louvain algorithm and an enhanced particle swarm optimization (PSO) algorithm for HEA community detection.

Main Methods:

• Development of interaction networks based on 6 descriptors for 90 high-entropy alloys (HEAs).
• Application of two community detection algorithms: the Louvain algorithm and an enhanced particle swarm optimization (PSO) algorithm.
• Validation of community detection accuracy using modularity metrics.

Main Results:

• Identification of 13 distinct alloy communities within the HEA dataset.
• The PSO-based community detection algorithm demonstrated an average accuracy improvement of 0.26 over the Louvain algorithm.
• Prediction of HEA characteristics, such as phase composition, with approximately 93% precision using the extracted communities.

Conclusions:

• Interaction network analysis is a viable approach for discovering functional communities in high-entropy alloys.
• The enhanced PSO algorithm offers superior performance for HEA community detection compared to the Louvain algorithm.
• This method facilitates the prediction of alloy properties and the discovery of new HEAs, reducing the need for extensive experimentation.