EUVIMEDEUVIMED entdecken
Crossref

Accurate Bandgap Prediction of Nitride‐Based Semiconductors via Stacking Ensemble Learning, Feature Engineering, and Dimensionality Reduction

Elham Zare, Fardin Taghizadeh, Faramarz Kanjouri

physica status solidi (b) · 2026-11 · 263 (11)

DOI: 10.1002/pssb.70359 ↗

Kurz gefasst

Fachartikel in der Zeitschrift physica status solidi (b), online erschienen am 17.11.2026. Verfasst von Elham Zare und 2 weiteren Autor:innen. Die Arbeit nennt 28 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Machine learning techniques were employed to develop predictive models for the HSE06 bandgaps of 300 nitride‐based compounds using fundamental atomic descriptors and GGA‐PBE bandgaps as input features. To further enhance the accuracy of bandgap prediction, a stacking framework was implemented by integrating the outputs of multiple baseline models. Four machine learning algorithms, including random forest (RF), light gradient boosting (LGB), Extreme gradient boosting (XGB), and gradient boosting decision tree (GBDT), were optimized for bandgap regression. Subsequently, the predictions generated by the best‐performing baseline models were used as input features for the stacking model. The predictive performance of all models was systematically evaluated using RMSE, MAE, MAPE, and R 2 metrics. Among all investigated approaches, the stacking model exhibited the highest predictive capability, achieving R 2 = 0.9493, RMSE = 0.1211, MAE = 0.0613, and MAPE = 0.0278. In addition, feature engineering was conducted on the atomic descriptors to further improve model performance. Seven different feature categories, namely atomic features (AF), atomic difference features (ADF), statistical features (SF), AF + SF, AF + ADF, ADF + SF, and AF + ADF + SF, were examined as predictor sets. The ADF and AF + SF feature groups demonstrated the most accurate predictions, indicating the strong correlation between bandgap characteristics and these descriptor classes. Furthermore, three stages of dimensionality reduction were applied to the feature sets, and the resulting model performances were systematically analyzed and compared. Overall, the obtained results convincingly demonstrate the outstanding effectiveness and robustness of the stacking strategy for accurate bandgap regression of nitride‐based materials.

Zitieren

APA

Zare, E., Taghizadeh, F., & Kanjouri, F. (2026). Accurate Bandgap Prediction of Nitride‐Based Semiconductors via Stacking Ensemble Learning, Feature Engineering, and Dimensionality Reduction. physica status solidi (b), 263(11). https://doi.org/10.1002/pssb.70359

Vancouver

Zare E, Taghizadeh F, Kanjouri F. Accurate Bandgap Prediction of Nitride‐Based Semiconductors via Stacking Ensemble Learning, Feature Engineering, and Dimensionality Reduction. physica status solidi (b). 2026;263(11). doi:10.1002/pssb.70359

BibTeX
@article{Zare2026,
  author = {Zare, Elham and Taghizadeh, Fardin and Kanjouri, Faramarz},
  title = {Accurate Bandgap Prediction of Nitride‐Based Semiconductors via Stacking Ensemble Learning, Feature Engineering, and Dimensionality Reduction},
  journal = {physica status solidi (b)},
  year = {2026},
  volume = {263},
  number = {11},
  doi = {10.1002/pssb.70359}
}
Kostenlos starten →