Objective: Aiming at the problems of clinical pathway optimization and medical cost control for stroke patients, this study proposes a clinical pathway discrimination model based on the XGBoost integrated learning algorithm. Combined with the multi-objective programming (MOP) method, this study explores its application value under the Diagnosis-Intervention Packet (DIP) payment model. Methods: The data of stroke patients (ICD codes I60–I63) from 2018 to 2022 were obtained through the medical record statistical management system of medical institutions in Guangdong Province. Efficiency indicators (average length of hospital stay), health economics indicators (total cost), effectiveness indicators (mortality rate), and specific indicators (NIHSS score) were extracted. The XGBoost algorithm was used to construct the clinical pathway discrimination model, and the hyperparameters were optimized through grid search. Based on the DIP payment rules, the prediction results of the model were used as inputs to establish a dynamic programming model, aiming to minimize costs and maximize curative effects to obtain the optimal clinical pathway plan. Results: The goodness-of-fit (R ²) of the XGBoost model on the test set reached 0.768, which was significantly better than that of the random forest (0.691) and the BP neural network (0.343). The total cost of the clinical pathway optimized by the dynamic programming model decreased by 12.7% (95% CI: 10.2–15.1%), the average length of hospital stay was shortened by 1.8 days, and the NIHSS improvement rate increased by 8.3%. Conclusion: The integrated model proposed in this study has high accuracy and robustness in clinical pathway discrimination. Combined with the MOP method, it can provide a scientific basis for the optimization of medical resources under the DIP model, helping medical institutions achieve the dual goals of precise cost control and curative effect improvement.