The study analyzed 894 patients with tibial plateau fractures (TPF), of which 299 (33.4%) had preoperative deep vein thrombosis (DVT) confirmed by duplex ultrasonography. They developed an interpretable machine learning diagnostic model based on routine preoperative variables, selecting nine key predictors: D-dimer, age, erythrocyte sedimentation rate, prognostic nutritional index, C-reactive protein, lymphocyte count, Schatzker type, neutrophil count, and smoking. The best results were achieved by the XGBoost algorithm with an AUROC of 0.840 (95% CI 0.790–0.884), an accuracy of 0.787, a sensitivity of 0.640, a specificity of 0.860, an F1 score of 0.667, and a Brier score of 0.149. SHAP analysis showed that D-dimer and age have a dominant non-linear effect; higher C-reactive protein, erythrocyte sedimentation, advanced Schatzker types and smoking increase the risk of DVT, while a lower prognostic nutritional index. The model provided a net clinical benefit according to decision curve analysis. It requires external validation in multicenter studies with recalibration and impact assessment.