Machine learning-based prediction of perovskite material photovoltaic conversion efficiency grades

The significance of perovskite materials in the field of renewable energy is becoming increasingly prominent, with their photovoltaic conversion efficiency emerging as the focal point of research. Traditional research and assessment methods are not only time-consuming but also labor-intensive. However, in recent years, machine learning technologies have provided new insights into addressing such challenges. Drawing upon a review of prior research, this paper categorizes the photovoltaic conversion efficiency of perovskite materials into high, medium, and low grades. Subsequently, we employed six mainstream machine learning algorithms, including Logistic Regression, Support Vector Machine (SVM), Decision Tree (DT), AdaBoost, XGBoost, and Long Short-Term Memory (LSTM) for training and prediction. Comprehensive evaluations revealed that SVM outperformed in overall accuracy, while XGBoost showcased superior classification performance based on the AUROC metric. Notably, the DT model excelled in terms of the F1 score. In contrast, both Logistic Regression and LSTM showed slightly inferior performance across various evaluation metrics, particularly in the F1 score and AUC. This study not only furnishes an efficient evaluation methodology for the photovoltaic conversion efficiency of perovskite materials but also offers valuable insights for research in related fields. We anticipate that as technology advances, we can provide more accurate and efficient solutions for real-world applications.