<p>Seismic attributes encapsulate substantial reservoir characterization information and can effectively support reservoir prediction. Given the high-dimensional nonlinear between sandbodies and seismic attributes, this study employs the RFECV method for seismic attribute selection, inputting the optimized attributes into a LightGBM model to enhance spatial delineation of sandbody identification. By constructing training datasets based on optimized seismic attributes and well logs, followed by class imbalance correction as input variables for machine learning models, with sandbody probability as the output variable, and employing grid search to optimize model parameters, a high-precision sandbody prediction model was established. Taking the 3D seismic data of Block F3 in the North Sea of Holland as an example, this method successfully depicted the three-dimensional spatial distribution of target formation sandstones. The results indicate that even under strong noise conditions, the multi-attribute sandbody identification method based on LightGBM effectively characterizes the distribution features of sandbodies. Compared to unselected attributes, the prediction results using selected attributes have higher vertical resolution and inter-well conformity, with the prediction accuracy for single wells reaching 80.77%, significantly improving the accuracy of sandbody boundary delineation.</p>