Evidence record · UDINUS · SDGs
- Unit
- Rektorat UDINUS
- Period
- 2026
- Goals
- Record
- d3f9e20d-7eb7-4ef2-a79e-076f1a6fdff1
- Published
- 2026-10-30
Universitas Dian Nuswantoro (UDINUS), through the participation of Prabowo, Wahyu Aji Eko; Akrom, Muhamad; Rustad, Supriadi; Herowati, Wise; Azies, Harun Al; Dipojono, Hermawan Kresno; Subagjo; Rusydi, Febdian; and Kasai, Hideaki, conducted research on ensemble learning for classifying and predicting the pore-limiting diameter (PLD) of metal-organic frameworks (MOFs). MOFs are tunable porous materials that have been widely studied for gas adsorption, separation, and catalysis. The activity addressed PLD classification and prediction for rational MOF screening and design.
The research developed an ensemble machine-learning pipeline using Random Forest (RF), XGBoost (XGB), and Light Gradient Boosting Machine (LGBM). The pipeline classified pore classes and performed PLD regression with a curated dataset derived from the Cambridge Structural Database. To address class imbalance across pore classes, the researchers evaluated TomekLinks, Synthetic Minority Oversampling Technique (SMOTE), and SMOTETomek. These resampling strategies were assessed in relation to classification and regression performance.
Among the evaluated approaches, RF provided the strongest overall performance, while the RF and TomekLinks configuration delivered the best balance between classification and regression. On the held-out test set, the best model achieved an R2 of up to 0.999 and an RMSE as low as 0.124 Å for PLD regression. Feature-importance analysis indicated that metal-center elemental properties and linker-related topological descriptors both contributed substantially to PLD prediction.
The activity also included applicability-domain assessment using leverage/Williams-plot analysis. Most test structures were within the modeled domain, and out-of-domain cases were explicitly flagged. The proposed workflow enabled domain-aware PLD estimation for high-throughput MOF screening. The source states that external validation using independent MOF collections and uncertainty quantification remain important for broader deployment. The documented activity therefore demonstrates UDINUS participation in computational research on MOF classification, prediction, and screening without extending the findings beyond the reported dataset and analyses.