Predicting Antimicrobial Resistance Without Local Data: Zero‑Shot Learning in Kazakhstan

This repository contains the trained models, calibration objects, inference datasets, and analysis code for the paper. The study develops a zero-shot country-level transfer framework that predicts antimicrobial resistance (AMR) probabilities in Kazakhstan using global surveillance data and publicly available macroeconomic and antibiotic consumption features, without any Kazakhstani isolates in training. The primary result is a two-model calibrated ensemble (post-Soviet cluster + global LOCO) applied to six WHO-priority pathogens across 2017--2023.