Machine Learning-Assisted Prediction and Control of Bandgap for Organic-Inorganic Metal Halide Perovskites

Area of Science:

• Materials Science
• Computational Chemistry
• Machine Learning

Background:

• Perovskite materials offer tunable band gaps for diverse applications.
• Machine learning accelerates the discovery of new materials.
• Data noise in datasets hinders traditional predictive models.

Purpose of the Study:

• To develop an outlier removal strategy for enhancing machine learning model generalization.
• To determine the optimal configuration for predictive modeling of perovskite band gaps.
• To identify key chemical composition factors influencing perovskite band gaps.

Main Methods:

• Implemented an outlier removal strategy to assess its impact on model performance.
• Employed the gradient boosting regression tree (GBRT) algorithm for prediction.
• Utilized the Shapley Additive Explanation (SHAP) method to interpret model predictions.

Main Results:

• The GBRT algorithm achieved high accuracy with MAE of 0.0287, MSE of 0.0014, RMSE of 0.0377, and R-squared of 0.979.
• The SHAP analysis revealed that the ratio of Iodine (I) significantly impacts the band gap, followed by ratios of Lead (Pb), Bromine (Br), and Tin (Sn).
• Experimental validation confirmed element ratio stability bounds crucial for band gap control.

Conclusions:

• An effective outlier removal strategy enhances the accuracy of machine learning models for perovskite band gap prediction.
• The GBRT algorithm demonstrates superior performance in predicting perovskite band gaps.
• Understanding the influence of chemical composition ratios, particularly Iodine, is critical for precise band gap engineering in perovskites.