Objective: Since High Altitude Pulmonary Edema (HAPE) has a rapid onset and fast progression-posing serious health risks-and research on HAPE prediction remains limited, this study employs an ensemble learning and resampling approach to develop a HAPE risk prediction model, aiming to enhance predictive efficiency. Methods: The HAPE dataset was constructed based on 674 subjects, covering demographic data, vital signs and biochemical indicators. Under-sampling, Synthetic Minority Over-sampling Technique (SMOTE), Adaptive Synthetic Sampling (ADASYN), and a combination of SMOTE and Edited Nearest Neighbors (SMOTE+ENN), were applied to address data imbalance. Stacking generalization was employed to develop an ensemble prediction model (MLP+GB+RF-Logistic), using Multilayer Perceptron (MLP), Gradient Boosting (GB), and Random Forest (RF) as base learners, with Logistic Regression serving as the meta-learner, for predicting the risk of HAPE. Results: 46 clinical features associated with HAPE, such as neutrophils, leukocytes, and monocyte percentage, were identified. Among the resampling methods, SMOTE+ENN showed the most improvement. For the ensemble models, the MLP+GB+RF-Logistic model achieved the highest AUC and Recall, with an AUC value of \(0.926 \pm 0.015\) . Furthermore, model interpretability was analyzed using the SHapley Additive exPlanations (SHAP) algorithm, identifying the top influential clinical features contributing to the prediction. Conclusions: This study developed an efficient ensemble learning model for HAPE risk prediction and identified key clinical features that markedly influence model performance. These findings provide a robust and interpretable basis for improving clinical decision-making related to HAPE.