Researchers developed a machine learning model to predict prolonged healing of diabetic wounds beyond 8 weeks using dynamic changes in C-reactive protein (CRP). The study included 465 patients with type 2 diabetes and wounds 5–8 cm² in size, treated with debridement alone; the training set had 325 patients (2021–2024) and the validation set had 140 patients (2025). CRP was measured three times: on admission (CRP), preoperatively after antibiotics (CRP2nd), and postoperatively at discharge (CRP3rd); the percentage changes of these values showed the effectiveness of the anti-inflammatory treatment. LASSO regression selected 15 key variables including CRP2nd, CRP3rd and albumin (ALB). The best GradientBoosting model achieved an accuracy of 0.9357, a sensitivity of 0.8689, and a specificity of 0.9873 in the validation set. SHAP analysis showed CRP2nd to be the most influential predictor (mean absolute SHAP value 0.460); higher CRP2nd and CRP3rd predict longer healing, while higher albumin and favorable CRP changes protect it. The model is used for clinical risk stratification in diabetic wounds.