Machine Learning-Assisted Discovery of Lewis Base Additives for Defect Passivation in Perovskite Solar Cells

• • LightGBM model achieved 87% classification accuracy in predicting Lewis base additives that enhance PCE by ≥2%, enabling high-throughput screening from 146 experimental data points. • • SHAP analysis identified critical design rules: HOMO energy between −7.5 and −6.3 eV, additive concentration between 2.5 and 6.5 mg/mL, and molecular backbones with ≤2 oxygen atoms and <5 carbon atoms, providing quantitative guidelines for passivator design. • • Experimental validation: (S)-PCA and MCPD (Class Ⅱ) improved PCE by 2.22% and 2.01%, respectively, while 3-HMBN (Class Ⅰ) showed minimal gain, confirming the ML model's predictive power. • • The champion (S)-PCA device achieved a certified PCE of 24.05%, demonstrating the practical impact of ML-guided discovery in pushing perovskite solar cell efficiency toward theoretical limits.