Machine Learning Research Yields Breakthrough in Materials Design and Discovery

Strategic Insights -

Researchers at Shanghai University have proposed a domain knowledge-assisted data anomaly detection (DKA-DAD) workflow to improve the accuracy of machine learning models in materials design and discovery. The study, funded by the National Natural Science Foundation of China and the National Key Research & Development Program of China, found that the DKA-DAD approach achieved a 12% F1-score improvement in anomaly detection accuracy compared to a purely data-driven approach. The research also showed that the ML models trained on materials datasets processed through DKA exhibited an average 9.6% improvement in R2 for property prediction.

Key Takeaways:

  • The researchers proposed a domain knowledge-assisted data anomaly detection (DKA-DAD) workflow to improve machine learning model accuracy in materials design and discovery.
  • The DKA-DAD approach achieved a 12% F1-score improvement in anomaly detection accuracy compared to a purely data-driven approach.
  • The ML models trained on materials datasets processed through DKA exhibited an average 9.6% improvement in R2 for property prediction.
  • The study was funded by the National Natural Science Foundation of China and the National Key Research & Development Program of China.
  • The research used 180 synthetic datasets by injecting noise into 60 structured materials datasets collected from materials ML studies.
  • The study's lead author is Siqi Shi from Shanghai University, and the research involved a team of authors including Yue Liu, Shuchang Ma, Zhengwei Yang, Duo Wu, Yali Zhao, and Maxim Avdeev.

Statistics:

  • 12% F1-score improvement in anomaly detection accuracy achieved by the DKA-DAD approach.
  • 9.6% average improvement in R2 for property prediction exhibited by ML models trained on materials datasets processed through DKA.
  • 180 synthetic datasets constructed by injecting noise into 60 structured materials datasets collected from materials ML studies.
  • 60 structured materials datasets used in the study were collected from materials ML studies.

Sources:

  • NewsRx. Data on Machine Learning Detailed by Researchers at Shanghai University (Domain Knowledge-assisted Materials Data Anomaly Detection Towards Constructing High-performance Machine Learning Models). Information Technology Newsweekly. November 4, 2025; p 107.
  • Shi, S., Liu, Y., Ma, S., Yang, Z., Wu, D., Zhao, Y., & Avdeev, M. (2025). Domain knowledge-assisted materials data anomaly detection towards constructing high-performance machine learning models. Journal of Materiomics, 11(6), 2025.