Fungal infections (FIs) are common and severe complications in hospitalized patients, with early identification of high‐risk patients being critical for improving clinical outcomes. Although previous studies have attempted to predict FI using traditional methods, the accuracy and clinical applicability of existing models still need improvement. The aim of this study is to develop and validate an interpretable machine learning (ML) model for predicting FI risk in hospitalized patients, enhancing the precision of clinical decision‐making.

We utilized real‐world inpatient data from Zhongnan Hospital of Wuhan University. Feature selection was performed using least absolute shrinkage and selection operator, random forest, and support vector machine recursive feature elimination methods. We then developed ML models, including logistic regression, random forest, decision tree, extreme gradient boosting, and light gradient boosting machine, K‐nearest neighbors and histogram gradient boosting trees, which were validated on both the training set and an internal testing set. During the model development process, the Shapley additive explanations (SHAPs) method was employed to enhance the clinical interpretability of the models. Additionally, an online prediction tool based on Streamlit was developed, integrating the model into the clinical workflow.

The study cohort comprised 806,761 hospitalized patients after data cleaning. Baseline characteristics were well balanced between the two cohorts (p > 0.05), including demographics, disease types, and key clinical variables. The extreme gradient boosting model demonstrated excellent predictive performance, with an area under the curve of 0.952 (0.951–0.953) on the training cohort, and 0.952 (0.951–0.952) on the testing cohort, showing good internal validity. The model also exhibited favorable sensitivity, specificity, and accuracy (F₁ score: 0.605). SHAP analysis revealed key predictive features such as antimicrobial use, mechanical ventilation, and bacterial infections, which contributed significantly to the model's predictions. An interactive Streamlit‐based web tool was developed, providing real‐time risk assessment, SHAP‐based interpretability, and risk‐stratified clinical recommendations.

This study successfully developed and validated a ML‐based model for FI prediction, demonstrating good predictive accuracy and strong clinical interpretability. Given its performance characteristics, the model is well‐suited for targeted clinical scenarios where confirming high‐risk patients is the priority, rather than for broad universal screening. Within this appropriate context, the model can assist clinicians in early identification of high‐risk patients and in making personalized intervention decisions, holding significant clinical value.