Explainable fault classification mechanism using solar data based on ensemble paradigm for efficient hydrogen energy production

The efficiency of green hydrogen production, which primarily relies on solar energy, depends on the optimal operation of the solar panels. Traditional machine learning (ML) approaches for solar panel fault classification face critical limitations, including poor performance on imbalanced datasets and limited transparency in decision-making. These methods tend to favor majority classes, leading to the misclassification of rare but essential fault types, which can significantly reduce the efficiency of hydrogen production. This work addresses these challenges by introducing an Explainable Ensemble Fault Classification (EEFC) model that integrates multiple classifiers through a voting-based ensemble strategy. For comprehensive evaluation, the EEFC model is compared not only with individual learners (GB, CB, XGBoost, LGBM, and RF) but also with other ensemble approaches such as stacking and blending, ensuring a fair benchmark. Experimental results on highly imbalanced and resampled datasets (SMOTE and ADASYN) demonstrate that EEFC consistently outperforms all baselines, achieving the highest F1-score (0.893) and MCC (0.869) under ADASYN, indicating superior balanced classification performance. To further validate its generalizability, the proposed EEFC model was also tested on an open-source wind turbine dataset, where it maintained strong performance across F1-score and MCC. Furthermore, the integration of Explainable Artificial Intelligence (XAI) techniques, specifically SHapley Additive exPlanations (SHAP), provides model interpretability by identifying key features contributing to fault detection and classification. To ensure comprehensive evaluation, an ablation study is carried out to further validates the contributions of individual components in the EEFC framework. The proposed model enhances solar panel fault management, ensures more reliable hydrogen production, and improves the transparency and robustness of decision-making in renewable energy systems. The full implementation is available at Github . • EEFC model integrates multiple learners via voting for better fault detection and SHAP explainability. • Robust preprocessing pipeline generates synthetic samples to reduce overfitting and improve generalization. • Comparative evaluation shows EEFC outperforms stacking, blending, and standalone learners. • EEFC validated on solar and wind datasets, showing strong cross-domain fault classification. • Proposed framework enhances renewable energy reliability for efficient green hydrogen production.