Application of machine learning models in predicting the risk of thromboembolic events in patients with nonvariceal gastrointestinal bleeding

Figure 1 Study flow for the machine learning of internal and external validation set. CatBoost: Categorical boosting; LassoLR: L1 regularized logistic regression; ML: Machine learning; NVGB: Nonvariceal gastrointestinal bleeding; RF: Random forest; SVM: Support vector machines; XGBoost: Extreme gradient boosting.

Figure 2 The area under the receiver operating characteristic curve of different machine learning models with the internal validation set and the external validation set. A: The area under the receiver operating characteristic curve of different machine learning models with the internal validation set; B: The area under the receiver operating characteristic curve of different machine learning L models with the external validation set. CatBoost: Categorical boosting; LassoLR: L1 regularized logistic regression; SVM: Support vector machines; XGBoost: Extreme gradient boosting.

Figure 3 The calibration curve of different machine learning models. A: Categorical boosting, L1 regularized logistic regression; B: Random forest for predicting thromboembolic events showed good performance. CatBoost: Categorical boosting; LassoLR: L1 regularized logistic regression; SVM: Support vector machines; XGBoost: Extreme gradient boosting.

Figure 4 The decision curve analysis of different machine learning models. The results showed that compared with the D-dimer, all machine learning models demonstrated higher net benefits across most threshold probability ranges. CatBoost: Categorical boosting; LassoLR: L1 regularized logistic regression; SVM: Support vector machines; XGBoost: Extreme gradient boosting.

Figure 5 The SHapley Additive exPlanations summary plot. The mean SHapley Additive exPlanations values of 10 important features and the influence of input features on the model's output. ICU: Intensive care unit.