Machine learning-based prediction of severe

Background: Methods: Clinical data from 123 patients in the mild MPP group and 284 patients in the severe MPP group were analyzed retrospectively. The least absolute shrinkage and selection operator (LASSO) was applied to identify key clinical variables. Eight ML algorithms were compared, and the model with the highest area under the curve (AUC) was selected for risk prediction. SHapley Additive exPlanations (SHAP) were used to interpret the model results. Results: The random forest (RF) model demonstrated optimal performance for predicting the individual risk of severe MPP, with an accuracy of 0.92, specificity of 0.91, recall of 0.91, and F1 score of 0.91. SHAP analysis indicated that the factors most strongly associated with severe MPP included length of hospital stay, month of admission, and serum creatinine (Cr) levels, among others. Conclusions: Eight ML models were developed to predict the individual risk of severe MPP. The RF model exhibited superior performance among the algorithms tested. This approach enabled accurate prediction of severe MPP in pediatric patients and facilitated the identification of key clinical factors associated with disease severity.