Abstract: Strong heterogeneity and ambiguous logging responses of interlayers in tight sandstone reservoirs of the LN Oilfield, Tarim Basin, hinder reliable interlayer identification. To address this issue, we propose an ensemble-learning method that integrates random forest (RF) and adaptive boosting (AdaBoost) for reservoir interlayer recognition. Based on core thin-section interpretations, a composite feature system is constructed by combining multiple logging parameters. Adaptive synthetic oversampling (ADASYN) is employed to enlarge the dataset, and RF hyperparameters are optimized via grid search before embedding the tuned RF into the AdaBoost framework, where weighted iterative boosting improves the stability and accuracy of recognition under small-sample conditions. Using logging data from seven wells, four models (RF–AdaBoost, AdaBoost, single RF, and K-nearest neighbors, KNN) are established and compared. Results show that the RF–AdaBoost model achieves an average F1-score (the harmonic mean of precision and recall) of 94.33% on the test set, and yields high accuracy with clear boundary delineation for three interlayer types, i.e., argillaceous, calcareous, and petrophysical interlayers. Furthermore, barrier effects are evaluated by integrating cross-well profiles with injection–production relationships. The identified interlayers are consistent with reservoir geological characteristics and production responses, indicating pronounced interlayer sealing and lateral diversion effects during oil–water migration. The proposed method is applicable to interlayer identification under non-cored conditions and can support well placement, injection–production optimization, and remaining-oil exploitation in tight reservoir development.