Organic photovoltaics (OPVs) offer a promising pathway toward low-cost, flexible, and solution-processable solar energy technologies; however, rational materials design remains challenging due to the complex and nonlinear relationships between molecular electronic structure and optoelectronic performance. In this study, machine learning models are developed to predict the spectral overlap of organic photovoltaic materials, a physically meaningful descriptor that quantifies the compatibility between molecular absorption and the solar spectrum. Using a large-scale OPV molecular dataset, multiple regression models linear regression (LR), support vector regression (SVR), random forest (RF), and gradient-boosted regression trees (GBRT) are systematically evaluated under a fivefold cross-validation framework. Among these, ensemble-based models demonstrate superior predictive accuracy and robustness. To move beyond purely predictive performance, explainable machine learning analysis based on SHapley Additive exPlanations (SHAP) is employed to uncover interpretable structure–property relationships. The SHAP results consistently identify frontier orbital energies and gap-related descriptors as dominant contributors to spectral overlap, while revealing clear directional dependencies and nonlinear effects. Overall, this work establishes an interpretable, data-driven framework that links molecular electronic descriptors to spectral overlap, offering a valuable tool for accelerated screening and rational design of high-performance organic photovoltaic materials.