Stacking Ensemble Learning for Clinical Antibiotic Resistance Prediction: A Comprehensive Study of Random Forest, XGBoost, and a Meta-Model Hybrid with Explainable AI on the ATLAS Surveillance Dataset

Antimicrobial resistance (AMR) is rapidly becoming one of the most serious threats to global health, and the need for smarter, faster clinical decision-making tools has never been greater. This research proposes a stacking ensemble machine learning framework that combines Random Forest and XGBoost as base learners, guided by a Logistic Regression meta-model, to predict whether a bacterial isolate will be resistant or susceptible to a given antibiotic. To ensure clinical trust and transparency, the study integrates SHAP-based explainability, revealing which patient and microbiological factors most influence each prediction. In terms of patient outcomes, this approach enables clinicians to identify the right antibiotic before treatment begins, reducing failure rates, prolonged hospital stays, and preventable deaths. For antibiotic stewardship, the model's interpretable outputs give prescribing teams concrete, data-driven evidence to move away from broad-spectrum empirical treatment toward targeted therapy directly curbing resistance propagation. From a public health perspective, training on the ATLAS dataset, which spans over 917,000 isolates from 83 countries collected across nearly two decades, means the model captures genuine global resistance trends and geographic variation, offering insights that can directly inform national AMR action plans and international surveillance policy. Finally, at the health systems level, the framework is designed to be lightweight and deployable in resource-constrained environments, giving hospitals and clinics particularly in low- and middle-income countries where AMR burden is heaviest a scalable, low-cost tool that reduces unnecessary treatment costs, improves resource allocation, and strengthens institutional preparedness against resistant pathogens. Altogether, this research sits at the intersection of machine learning, clinical microbiology, and public health, and the ATLAS dataset is uniquely positioned as the foundation that makes this scale of contribution possible.