Comparison of Machine Learning Models for Accurate Prediction of Power Conversion Efficiency in Perovskite Solar Cell

Perovskite solar cells (PSCs) have great potential for renewable energy due to their high effectiveness, adjustable optoelectronic parameters, and low production costs. However, accurately predicting the power conversion efficiency(PCE) and stability is challenging for its complex material and structural factors. In this work, we used a large-scale open-access dataset of over 26,000 perovskite devices, integrated with extra compositional and structural parameters to predict the power conversion efficiency of PSCs. We applied and compared six machine learning models: LR, XGBoost, GBR, RF, KNN, and SVR. Our methodology included thorough data preprocessing, feature engineering (including converting categorical features), and extensive cleaning. These steps significantly improved modelling accuracy. MAPE, RMSE, MAE & $R^{2}$ are evaluated as performance evaluation metrics. GBR and RF showed higher performance than the others. The GBR model achieved the best MAPE, RMSE and $R^{2}$ of $0.0886,0.8051,0.969$, respectively. This research shows that well-chosen input features, combined with strong ML algorithms and a validated high-accuracy benchmark, can effectively model PSC’s performance. This provides a powerful predictive tool for accelerating materials optimization.