The study created and validated a clinical prediction model for perioperative pneumonia in elderly patients with hip fractures and preoperative stroke based on 698 patients (244 with pneumonia, 454 without), divided into a 7:3 training and validation set. Independent risk factors were pulmonary hypertension, respiratory failure, chronic obstructive pulmonary disease, type of surgery, age, albumin, hemoglobin and BNP levels. The nomogram model achieved an AUC of 0.9203 in the training set and 0.7356 in the validation set, with good fit of the calibration curves and clinical utility in the intermediate risk range according to DCA. SHAP analysis confirmed albumin, hemoglobin, age and BNP as the main predictive variables. Among the 10 machine learning models, logistic regression and LDA performed best with an AUC of 0.743, accuracy of 0.712 and 0.708, respectively. All models showed recall above 0.680, precision 0.650–0.660 and high F1 scores. The model enables individualized risk assessment and early intervention.