<p>Perovskite solar cells (PSCs) have rapidly established themselves as a leading contender for next-generation photovoltaics, offering tunable band gaps, structural flexibility, and impressive power conversion efficiency (PCE). However, significant stability issues—such as moisture sensitivity, thermal instability, halide segregation, and ion migration—have persistently been observed, limiting commercial viability. Traditional methods such as Density Functional Theory (DFT), chemical substitution, and synthesis have been used, but those are computationally intensive and experimentally demanding processes. So, data-driven methodologies have been explored as powerful allies to experimental research via high-throughput screening. In this paper, a hierarchical ensemble machine learning (ML) framework named PerovLearn has been proposed for accurate band gap prediction of perovskite materials to expedite photovoltaic design. Twenty-two ML algorithms from six different families have been investigated using Root Mean Squared Error (RMSE), Coefficient of Determination (R²), and Pearson and Spearman correlation coefficients. Top-performing models from each model family have been identified, and then ensemble strategies have been applied to enhance the generalization and to improve predictive stability. It is found that blended averaging and stacked generalization model with a Random Forest (RF) meta-learner has attained near-experimental accuracy (RMSE = 0.008751, R² = 0.999176) in bandgap prediction. In the PerovLearn framework, SHapley Additive exPlanations (SHAP) based interpretability has been incorporated to unveil compositional influences on bandgap, bridging predictive power and scientific insight. The proposed PerovLearn framework has delivered a generalized, adaptable, scalable, interpretable, and high-fidelity approach for bandgap engineering, offering a robust data-driven pathway for accelerated design and optimization of perovskite photovoltaic materials.</p>