Evidence record · UDINUS · SDGs
- Unit
- Rektorat UDINUS
- Period
- 2024
- Goals
- Record
- 11c1bb3c-fe0a-44f9-8f62-e27d80a721cf
- Published
- 2026-10-30
UDINUS conducted a study addressing the challenge of small datasets in materials informatics. The activity involved Rustad, Supriadi; Akrom, Muhamad; Sutojo, Totok; and Dipojono, Hermawan Kresno as participants. The study introduced the Gaussian Mixture Model Virtual Sample Generation, or GMM-VSG, approach. This approach generated virtual samples to enhance feature correlation. The activity applied GMM-VSG to six small datasets and one large dataset involving 218 N-heterocyclic compounds.
The study assessed predictive performance using several machine-learning methods. For Random Forest, the reported R2 increased from 0.80 to 0.99, while RMSE decreased from 9.87 to 0.22. For Kernel Ridge, R2 increased from 0.70 to 0.99, and RMSE decreased from 10.08 to 0.83. KNN also improved, with R2 reported across a range from 0.74 to 0.90. The study further reported notable improvements for ANN and MLPNN.
These findings identify GMM-VSG as an approach for improving predictive performance in materials informatics and for advancing anti-corrosion material research. This evidence draft records the UDINUS activity, the stated method, the stated dataset scope, and the reported findings without extending the source beyond those details.