The study developed and validated a 28-day mortality predictive model based on the lactate dehydrogenase/albumin ratio (LAR) in patients with severe pneumonia. A retrospective cohort study included 491 patients admitted to the hospital from January to May 2020. LAR was significantly associated with the risk of 28-day mortality and identified as an independent risk factor. LAR demonstrated a high area under the curve (AUC) in predicting mortality with a non-linear relationship to risk. Using the Boruto algorithm and machine learning models, the random forest (RF) model had the best predictive performance. SHAP analysis confirmed the dominant role of LAR in the RF model. LAR is an effective biomarker for predicting 28-day mortality, especially in nonseptic patients, and increases the accuracy of risk assessment.