Predicting negative blood cultures from EHR using Orange Data Mining

This repository contains teaching material for a hands-on workshop on reproducing a clinical machine-learning study with Orange Data Mining.

The session uses a prepared teaching version of the Antibiotic Resistance Microbiology Dataset (ARMD) to reproduce, in a simplified and visual way, the main ideas of the Cultryx blood-culture prediction study. The goal is to help students inspect clinical tabular data, understand missingness and class imbalance, apply basic preprocessing, train machine-learning models, and interpret diagnostic metrics such as sensitivity, specificity, PPV and NPV.

The archive includes:

- a Jupyter notebook used as the workshop slide deck
- prepared Orange Data Mining workflows for each exercise
- student and teacher versions of the Orange workflows
- a final complete Orange workflow
- prepared train and holdout datasets formatted for Orange
- precomputed figures and model-result tables
- scripts used to rebuild the notebook, figures, workflows and results

The prepared dataset is smaller and formatted for teaching, so it is easier to use in Orange during a workshop. It is derived from the original ARMD database available in Dryad: https://doi.org/10.5061/dryad.jq2bvq8kp

Orange Data Mining is used as the main teaching tool because it allows students to build and understand machine-learning workflows visually, without needing to write Python code.