Bandgap prediction of metal halide perovskites using regression machine learning models

Organometal halide perovskites represent a type of nanomaterials, which are extensively used in solar cells, light-emitting diodes, detectors and memristors due to their outstanding optical, electrical and mechanical properties. Here, we use a dataset composed of 240 perovskites to train two machine learning models, ElasticNet and Isotonic Regression, able to predict the bandgaps.

The performance of our ML models is evaluated using Correlation coefficient, Mean Absolute Error (MAE), and Root Mean Square Error (RMSE). The lowest MAE of 0.09 eV is calculated for Cs-based perovskites from ElasticNet and Ten-fold cross-validation results. While the highest MAE of 0.34 eV was obtained for MA-based perovskites with Isotonic Regression.

Furthermore, a high correlation value of 0.98 between the DFT calculated and ML predicted results is observed. From the detailed comparative analysis, ElasticNet emerges as a prominent machine learning model for predicting the bandgap of metal halide perovskites more accurately and can also be further employed to predict the various properties of materials and their selection for different applications as well as to expand the investigation to other structures and organic molecules.