Feature Selection and SHAP-Based Interpretability in ML Models for AMR Prediction in Klebsiella Pneumoniae

Antimicrobial resistance (AMR) compromises the body’s ability to fight infections and poses significant risks for patients receiving medical care. The increasing spread of resistant organisms—including Staphylococcus aureus, Enterococcus spp., Klebsiella pneumoniae, and Pseudomonas aeruginosa—is a major clinical concern. Advances in the use of comprehensive medical datasets and predictive analytics now support the development of evidence-driven approaches to AMR management. In this study, a robust prediction pipeline is introduced, combining feature selection methods, multiple classifier algorithms, and class imbalance correction. The findings indicate that Random Forest and Gradient Boosting models deliver superior performance, while SHAP-based interpretation highlights tet(D), tRNA, and contigs as critical contributors to resistance, providing transparent, patient-level risk insights.