<p>Accurate prediction of single-well productivity is critical for the efficient and economic development of low-resistivity reservoirs, where complex pore structures, high irreducible water saturation, and strong mineral conductivity often lead to poor prediction performance using conventional approaches. Traditional data-driven models frequently neglect geological constraints and struggle to capture the highly nonlinear relationships between reservoir properties and production behavior. This study proposes a Dual-Model Ensemble with Geological Post-Processing (DM-EGPP) framework for oil productivity prediction in low-resistivity reservoirs. The framework integrates two optimized gradient-boosting regressors, XGBoost and LightGBM, through independent hyperparameter optimization and an ultra-fine weight search strategy. Geological prior knowledge is incorporated via a post-processing module that includes layer-type screening, porosity–permeability threshold constraints, and a five-level adaptive error correction mechanism to enhance physical consistency. The proposed model is validated using field data from a representative low-resistivity reservoir. Under small-sample conditions, the DM-EGPP framework achieves a coefficient of determination (R²) of 0.8595, significantly outperforming individual baseline models such as support vector regression, random forest, and neural networks. Comparative evaluation demonstrates notable improvements in prediction accuracy, stability, and geological plausibility after applying the geological post-processing constraints. The results indicate that the DM-EGPP framework provides a robust and interpretable solution for productivity prediction in low-resistivity reservoirs. The novelty of this work lies in the systematic integration of dual-model ensemble learning with a geology-driven post-processing strategy, offering a practical approach that bridges data-driven prediction accuracy and geological rationality for complex reservoir systems.</p>