| Titre : | Prediction of perovskite materials for photovoltaic applications using Data Mining on DFT calculations and experimental study | | Type de document : | document multimédia | | Auteurs : | Soundous Touati, Auteur ; Ali Benghia, Directeur de thèse ; Zoulikha Hebboul, Directeur de thèse | | Editeur : | Laghouat : Université Amar Telidji - Département des sciences de la matière | | Année de publication : | 2025 | | Importance : | 134 p. | | Note générale : | Option : Materials physics | | Langues : | Anglais | | Mots-clés : | ABX3 Perovskite Machine learning XGBoost algorithm Random Forest algorithm Band gap energy Formation energy Open Quantum Materials Database (OQMD) Materials project database | | Résumé : | The accelerating progress of machine learning (ML) provides significant opportunities to address the limitations of conventional materials discovery, especially within the chemically diverse and technologically important ABX₃ perovskite family. In this thesis, ML is combined with materials science to predict key physicochemical properties—formation energy, band gap energy, crystal system, space group, and lattice volume—using descriptors derived only from elemental and structural attributes. Avoiding reliance on computationally expensive DFT features allows the models to perform fast and scalable screening of novel compounds. Ensemble approaches, specifically Random Forest and XGBoost, were applied to datasets from the Materials Project and the Open Quantum Materials Database (OQMD). Evaluation metrics, such as R², MAE, and RMSE, confirmed the strong predictive capacity and generalizability of the models, demonstrating precision comparable to DFT-dependent studies. The methodology was extended to explore over 13,000 perovskite compounds, resulting in the identification of 1,836 novel candidates absent from OQMD with favorable stability and electronic characteristics.
This work makes three key contributions: (i) the development of a reliable ML framework for multi-property prediction, (ii) the establishment of a scalable discovery pipeline relying solely on chemical formulas, and (iii) new insights into the structural and electronic behavior of perovskites. Overall, the findings position ML as a transformative approach to rational material design and the accelerated discovery of functional compounds. | | note de thèses : | Thèse de doctorat en physique |
Prediction of perovskite materials for photovoltaic applications using Data Mining on DFT calculations and experimental study [document multimédia] / Soundous Touati, Auteur ; Ali Benghia, Directeur de thèse ; Zoulikha Hebboul, Directeur de thèse . - Laghouat : Université Amar Telidji - Département des sciences de la matière, 2025 . - 134 p. Option : Materials physics Langues : Anglais | Mots-clés : | ABX3 Perovskite Machine learning XGBoost algorithm Random Forest algorithm Band gap energy Formation energy Open Quantum Materials Database (OQMD) Materials project database | | Résumé : | The accelerating progress of machine learning (ML) provides significant opportunities to address the limitations of conventional materials discovery, especially within the chemically diverse and technologically important ABX₃ perovskite family. In this thesis, ML is combined with materials science to predict key physicochemical properties—formation energy, band gap energy, crystal system, space group, and lattice volume—using descriptors derived only from elemental and structural attributes. Avoiding reliance on computationally expensive DFT features allows the models to perform fast and scalable screening of novel compounds. Ensemble approaches, specifically Random Forest and XGBoost, were applied to datasets from the Materials Project and the Open Quantum Materials Database (OQMD). Evaluation metrics, such as R², MAE, and RMSE, confirmed the strong predictive capacity and generalizability of the models, demonstrating precision comparable to DFT-dependent studies. The methodology was extended to explore over 13,000 perovskite compounds, resulting in the identification of 1,836 novel candidates absent from OQMD with favorable stability and electronic characteristics.
This work makes three key contributions: (i) the development of a reliable ML framework for multi-property prediction, (ii) the establishment of a scalable discovery pipeline relying solely on chemical formulas, and (iii) new insights into the structural and electronic behavior of perovskites. Overall, the findings position ML as a transformative approach to rational material design and the accelerated discovery of functional compounds. | | note de thèses : | Thèse de doctorat en physique |
|
Prediction of perovskite materials for photovoltaic applications using Data Mining on DFT calculations and experimental study
