A study evaluated prediction models for postoperative delirium (POD) in elderly patients with hip fractures using conventional logistic regression and five machine-learning algorithms. The study included 797 patients aged ≥65 years who underwent hip fracture surgery between 2014 and 2019. The incidence of POD was 9.28%. The logistic regression model identified age, renal insufficiency, COPD, antipsychotics, LDH, and CRP as risk factors, with an AUC of 0.71. Machine-learning models, including Random Forest (AUC 0.81), Gradient Boosting Machine (AUC 0.77), and XGBoost (AUC 0.77), showed higher accuracy. The Random Forest model had the highest sensitivity (91.9%) and precision (67.8%). SHAP values were used to interpret feature importance, highlighting BNP, Troponin T, CRP, and CK-MB as key predictors. The study demonstrated that machine-learning models offer better risk stratification for POD, improving outcomes for elderly hip fracture patients. The results suggest that focusing on modifiable biomarkers can help reduce POD risk. The study had limitations, including a retrospective design and single-center validation. Overall, the models provide valuable tools for clinicians to identify high-risk patients and improve care.A study evaluated prediction models for postoperative delirium (POD) in elderly patients with hip fractures using conventional logistic regression and five machine-learning algorithms. The study included 797 patients aged ≥65 years who underwent hip fracture surgery between 2014 and 2019. The incidence of POD was 9.28%. The logistic regression model identified age, renal insufficiency, COPD, antipsychotics, LDH, and CRP as risk factors, with an AUC of 0.71. Machine-learning models, including Random Forest (AUC 0.81), Gradient Boosting Machine (AUC 0.77), and XGBoost (AUC 0.77), showed higher accuracy. The Random Forest model had the highest sensitivity (91.9%) and precision (67.8%). SHAP values were used to interpret feature importance, highlighting BNP, Troponin T, CRP, and CK-MB as key predictors. The study demonstrated that machine-learning models offer better risk stratification for POD, improving outcomes for elderly hip fracture patients. The results suggest that focusing on modifiable biomarkers can help reduce POD risk. The study had limitations, including a retrospective design and single-center validation. Overall, the models provide valuable tools for clinicians to identify high-risk patients and improve care.