XAS Data Preprocessing of Nanocatalysts for Machine Learning Applications.

Area of Science:

• Materials Science
• Chemistry
• Data Science

Background:

• Nanocatalyst development relies on precise characterization, often generating large, heterogeneous datasets.
• Extracting knowledge from experimental data, particularly X-ray absorption spectroscopy (XAS), is key to accelerating materials discovery.
• Current research often overlooks the critical XAS data preprocessing stage for machine learning (ML) applications.

Purpose of the Study:

• To develop automated tools for preprocessing and presenting physical experimental data, specifically XAS data.
• To create deposited datasets for palladium-based nanocatalysts to facilitate ML-driven research.
• To bridge the gap between raw XAS data acquisition and its application in ML models.

Main Methods:

• Developed a software toolkit implementing a single pipeline for XAS data preprocessing.
• Utilized principal component analysis (PCA), z-score normalization, and the interquartile method for data cleaning.
• Applied k-means clustering for material phase identification based on experimental feature vectors.

Main Results:

• Successfully created a pipeline for automated XAS data preprocessing.
• Generated deposited datasets of physical experiments on palladium-based nanocatalysts.
• Demonstrated efficient data preparation for ML analysis and prediction.

Conclusions:

• Automated preprocessing of XAS data is essential for reliable ML analysis in nanocatalyst research.
• The developed tools and datasets will accelerate knowledge extraction and materials discovery.
• Facilitates data dissemination and reuse among researchers for advancing artificial intelligence in materials science.