DFT-Guided Machine Learning and Device Modeling of Cs2PtBr6 Lead-Free Perovskite Solar Cells

DFT:1. The structure of the material was optimized using the CASTEP module, parameters contained in parameters.txt file2. Optimized structure was then simulated again to find electronic, elastic and optical properties using their respective parameters files.3. The results are stored in Electronic, Optics and Elastic Constants files.4. The csv files for each properties were prepared from these files and the data was plotted via OriginPro softwareML:To train the models from scratch, open perovskite20kfinal.ipynb either in your local Jupyter Notebook or on Kaggle/Google Colab (recommended, since the required libraries are already pre-installed). 1. The Excel file pero_20k_abs.xlsx should be loaded from the Datasets folder in the second code block titled “Exporting the Excel file as Pandas DataFrame”.2. All code blocks should be executed sequentially so that the models are trained and saved.3. The test dataset real_etl_htl_with_abs.csv should be loaded, and predictions should be performed using the newly trained models.4. The outputs of the workflow were stored and organized in the following manner: * The models were trained using the scikit-learn framework, and their learning progress along with accuracy metrics were saved. * The real vs. predicted plots were generated for PCE, Jsc, Voc, and FF to evaluate predictive performance. * Feature importance maps, SHAP plots, and correlation visualizations were produced to highlight the contribution of input features. * Error histograms were created to compare training and test errors. * The final predictions on the real dataset, as well as summary tables of accuracy, were stored in the Prediction Results folder. * All datasets, model files, prediction outputs, and visualizations were systematically maintained in the documented folder hierarchy. root│├── perovskite20kfinal.ipynb - Main Jupyter Notebook for training and evaluating models├── Datasets/│ ├── real_etl_htl_with_abs.csv - Test dataset containing real ETL, HTL, and absorber data for model predictions│ └── pero_20k_abs.xlsx - Training and validation dataset with 20,000 simulated absorber layer combinations│├── Models/│ ├── Voc_xgb_model_multi_stage_abs.pkl - Trained XGBoost model for Voc prediction│ ├── Voc_rf_model_multi_stage_abs.pkl - Trained Random Forest model for Voc prediction│ ├── Voc_mlp_scaler.pkl - Data scaler used for MLP Voc model│ ├── Voc_mlp_model.h5 - Trained MLP model for Voc prediction│ ├── Voc_mlp_metrics.pkl - Voc MLP model saved with metrics│ ├── Voc_knn_model_abs.pkl - Trained KNN model for Voc prediction│ ├── Voc_histgbr_model_abs.pkl - Trained Histogram Gradient Boosting model for Voc prediction││ ├── PCE_xgb_model_multi_stage_abs.pkl - Trained XGBoost model for PCE prediction│ ├── PCE_rf_model_multi_stage_abs.pkl - Trained Random Forest model for PCE prediction│ ├── PCE_mlp_scaler.pkl - Data scaler used for MLP PCE model│ ├── PCE_mlp_model.h5 - Trained MLP model for PCE prediction│ ├── PCE_mlp_metrics.pkl - PCE MLP model saved with metrics│ ├── PCE_knn_model_abs.pkl - Trained KNN model for PCE prediction│ ├── PCE_histgbr_model_abs.pkl - Trained Histogram Gradient Boosting model for PCE prediction││ ├── Jsc_xgb_model_multi_stage_abs.pkl - Trained XGBoost model for Jsc prediction│ ├── Jsc_rf_model_multi_stage_abs.pkl - Trained Random Forest model for Jsc prediction│ ├── Jsc_mlp_scaler.pkl - Data scaler used for MLP Jsc model│ ├── Jsc_mlp_model.h5 - Trained MLP model for Jsc prediction│ ├── Jsc_mlp_metrics.pkl - Jsc MLP model saved with metrics│ ├── Jsc_knn_model_abs.pkl - Trained KNN model for Jsc prediction│ ├── Jsc_histgbr_model_abs.pkl - Trained Histogram Gradient Boosting model for Jsc prediction││ ├── FF_xgb_model_multi_stage_abs.pkl - Trained XGBoost model for FF prediction│ ├── FF_rf_model_multi_stage_abs.pkl - Trained Random Forest model for FF prediction│ ├── FF_mlp_scaler.pkl - Data scaler used for MLP FF model│ ├── FF_mlp_model.h5 - Trained MLP model for FF prediction│ ├── FF_mlp_metrics.pkl - FF MLP model saved with metrics│ ├── FF_knn_model_abs.pkl - Trained KNN model for FF prediction│ └── FF_histgbr_model_abs.pkl - Trained Histogram Gradient Boosting model for FF prediction│├── Prediction Results/│ ├── top10pred.png - Visualization of the top 10 model predictions│ ├── targeted_pred.png - Visualization of targeted prediction results│ ├── model_accuracy_table.png - Table summarizing model accuracy scores│ └── df_real_with_predictions_for20k.csv - CSV containing predictions on real dataset using trained models│└── Viz/ ├── Training and Test Error Histogram XGBoost/ │ ├── PCE_xgb_training_error_hist.png - Histogram of training error distribution for the XGBoost PCE model │ └── PCE_xgb_test_error_hist.png - Histogram of test error distribution for the XGBoost PCE model │ ├── Real vs Predicted XGBoost/ │ ├── Voc_xgb_actual_vs_predicted.png - Scatter plot comparing actual vs. predicted Voc values using XGBoost │ ├── PCE_xgb_actual_vs_predicted (2).png - Scatter plot comparing actual vs. predicted PCE values using XGBoost │ ├── Jsc_xgb_actual_vs_predicted.png - Scatter plot comparing actual vs. predicted Jsc values using XGBoost │ ├── fig_grid (1).png - Grid layout visualization of multiple prediction result plots │ └── FF_xgb_actual_vs_predicted.png - Scatter plot comparing actual vs. predicted FF values using XGBoost │ ├── Feature Importance and Correlation XGBoost/ │ ├── Voc_xgb_model_multi_stage_abs_corr_importance_featmap.png - Feature importance and correlation heatmap for Voc prediction with XGBoost │ ├── SHAP_PCE.png - SHAP feature importance summary plot for PCE prediction │ ├── PCE_xgb_model_multi_stage_abs_corr_importance_featmap (1).png - Feature importance and correlation heatmap for PCE prediction with XGBoost │ ├── Jsc_xgb_model_multi_stage_abs_corr_importance_featmap.png - Feature importance and correlation heatmap for Jsc prediction with XGBoost │ └── FF_xgb_model_multi_stage_abs_corr_importance_featmap.png - Feature importance and correlation heatmap for FF prediction with XGBoost │ ├── PCEhist.png - Histogram showing the distribution of PCE values across the dataset └── ML FLOWCHART.png - Flowchart illustrating the machine learning pipeline used in the projectSCAPS-1D: 1. The SCAPS-1D simulation software was utilized for these computations using a Windows operating system. 2. Different parameter given from previous research. 3. Different parameter have been optimized.