A machine learning framework to predict the severity and management of pediatric appendicitis using clinical markers and five explainable techniques

Appendicitis remains a common surgical emergency in children and adolescents, where timely diagnosis and accurate severity assessment are critical to prevent complications such as perforation and peritonitis. In this study, we applied machine learning (ML) approaches to clinical data to enhance prediction of both disease severity and the need for surgical versus conservative management. Compared with traditional logistic regression, ML models demonstrated improved predictive performance, particularly in identifying high-risk patients. The most influential variables included length of hospital stay, loss of appetite, C-reactive protein (CRP), presence of peritonitis, and the Alvarado score. These predictors, when analyzed through ML, allowed better stratification of patients into severe versus non-severe categories, achieving 94 % accuracy for severity prediction and 89 % accuracy for management prediction. Explainable AI techniques further clarified the role of each variable, ensuring interpretability for clinicians. Our findings highlight that ML-based decision support tools can complement conventional clinical assessment, potentially reduce diagnostic delays and supporting more precise treatment planning in hospital settings.