Abstract: Tight sandstone gas reservoirs are currently a key focus in oil and gas exploration. However, their complex pore structures, low permeability, and high bound water content pose significant challenges for well logging evaluation. While Nuclear Magnetic Resonance (NMR) logging can accurately distinguish between bound and movable water in the reservoir, it is not available for all exploration wells. Therefore, predicting bound water saturation and calculating permeability from conventional logging data remains limited and requires further study. This paper proposes a stacked ensemble learning method for predicting bound water saturation and evaluating permeability in tight sandstone reservoirs. Bound water saturation calculated from NMR logging is used as the prediction label, while conventional logging data and reservoir characteristic parameters derived from rock physics models are used as input features. A stacked ensemble learning framework is constructed, integrating physical models with data-driven approaches , consisting of multiple base learners (Random Forest, Extreme Gradient Boosting, and Gradient Boosting) and a meta-learner (Random Forest). Initial predictions from the base learners are input into the meta-learner, which is trained using cross-validation to predict bound water saturation from conventional logs. Permeability is then calculated based on the predicted bound water saturation using a regressed Timur equation, combined with rock physics experimental results, to realize permeability estimation of tight sandstone gas-bearing reservoirs. This method was applied to the tight sandstone reservoirs of the HG Member in the XH Depression, located in the C Block of the East China Sea Shelf Basin. The mean absolute error of the predicted bound water saturation was less than 5%, and the logarithmic error in permeability was below 0.2. The proposed method significantly improves the accuracy of bound water saturation prediction and permeability evaluation in tight sandstone reservoirs and demonstrates strong adaptability and applicability for broader use.