The aim of the study is to develop an ensemble machine learning (ML) method that enables the construction of interpretable prognostic models and to test it using the example of predicting in-hospital mortality (IHM) in patients with ST-segment elevation myocardial infarction (STEMI). A retrospective cohort study was conducted using data from 4673 electronic medical records of patients with STEMI who underwent percutaneous coronary intervention (PCI). Two groups of individuals were identified, the first of which consisted of 318 (6.8%) patients who died in hospital, the second - 4355 (93.2%) - with a favorable outcome of PCI. Using multimetric categorization methods (minimizing p-value, maximizing the area under the ROC curve-AUC and shap-value analysis results), the predictors were transformed into risk factors (RF) for IHM. To develop prognostic models of IHM, multivariate logistic regression, random forest RF (RandFRF), stochastic gradient boosting, random forest, Adaptive boosting, Gradient Boosting, Light Gradient-Boosting Machine and CatBoost were used. The authors developed the RandFRF method, which aggregates the predictions of modified decision trees, identifies key risk factors (RFs), and ranks them based on their contribution to the likelihood of the adverse outcome (IHM). The RandFRF method demonstrated a high prognostic capacity (AUC = 0.897), comparable to that of XGBoost (AUC = 0.891), Random Forest (AUC = 0.887), and CatBoost (AUC = 0.881). Importantly, RandFRF also enables clinical interpretation of the model outputs by quantifying the influence of each risk factor on the predicted probability of IHM. This method may serve as a reliable tool for developing interpretable ML models in clinical medicine.