Abstract: Prediction of gas-bearing tight sandstone reservoirs is extremely challenging, and conventional methods struggle to address such issues effectively. Owing to the limitations of limited sample data and poor data quality, achieving accurate intelligent prediction using machine learning remains a significant challenge. To tackle these problems, this paper proposes a logging-based fluid prediction method for gas-bearing tight sandstone reservoirs based on the RTF-MR-stacking model. Specifically, the RTF model is first designed for data sample augmentation. Subsequently, the improved undersampling algorithm MAHAKIL combined with the average inter-class potential is employed to optimize the augmented samples. Finally, an enhanced stacking module is utilized to implement fluid identification. Experimental results demonstrate that the proposed method achieves an accuracy of 93% in fluid prediction for highly heterogeneous tight sandstones in the Sulige Gas Field. It effectively mitigates the adverse impact of limited sample data in the study area on reservoir fluid prediction, and its performance metrics outperform those of other mainstream machine learning models and similar stacking models. Validation results indicate that the fluid prediction model based on RTF-MR-stacking exhibits excellent applicability in the study area. It can provide high-quality technical support and theoretical guidance for productivity prediction of old and blind wells in the region, and also offers a new perspective for conventional fluid evaluation of tight sandstone reservoirs.