High-Accuracy Bandgap Prediction and Classification in Hybrid and Inorganic Halide Perovskites Using Advanced Machine Learning Techniques

Hybrid and inorganic halide perovskites (HP) have garnered significant attention for their applications in solar cells, LEDs, and sensors due to their exceptional electronic and optical properties. The accurate prediction and classification of bandgaps in these materials are crucial for advancing their technological potential. Traditional methods like Density Functional Theory (DFT) are computationally expensive, motivating the use of machine learning (ML) as a faster and more efficient alternative. In this study, we analyze 7382 hybrid and inorganic HP using a diverse set of ML models to classify materials based on whether they exhibit zero or nonzero bandgaps, and to predict their bandgap values. For regression tasks, AdaBoost Regressor (ABR), decision tree regressor (DTR), and gradient boosting regressor (GBR) were employed, while gradient boosting machines (GBM), decision tree (DT), and Multilayer Perceptron (MLP) were used for classification. Evaluation metrics for prediction included mean absolute error (MAE), mean squared error (MSE), and the R². For classification, metrics, such as accuracy, precision, recall, F1-score, area under the ROC curve (AUC-ROC), and area under the precision-recall curve (AUC-PR) were utilized. Results indicate that ABR achieved the highest prediction accuracy (MSE ≈ 0.074 eV, MAE ≈ 0.088 eV, R² ≈ 91.1% for direct bandgaps; MSE ≈ 0.041 eV, MAE ≈ 0.076 eV, R² ≈ 93.4% for indirect bandgaps). In classification, the GBM model outperformed others, achieving 96% and 97% accuracy for direct and indirect bandgaps, respectively. Feature analysis revealed that elemental properties, such as valence and group of constituent elements, particularly their mean and standard deviation, play a dominant role in bandgap determination. These findings highlight the potential of ML-driven approaches in accelerating perovskite material discovery and optimizing their electronic properties for future optoelectronic applications.