Quantum ML for ICU Mortality Prediction

A reproducible benchmark of Quantum SVMs, a VQC, and a QNN against classical baselines on the WiDS Datathon 2020 ICU dataset. Built on Qiskit, fully seeded, and honest about what it found.

View the findings Try the live demo Source on GitHub

Headline: at this scale, the classical models win cleanly and the quantum models sit at chance. Logistic Regression reaches ROC-AUC 0.817 [0.787, 0.845]; every quantum model's 95% confidence interval includes 0.5, while running two to four orders of magnitude slower. That negative result is the point: it is what a careful, leakage-free audit actually shows.

At a glance

--

Best classical ROC-AUC

 

--

Best quantum ROC-AUC

 

--

Models benchmarked

classical + quantum

--

Quantum slowdown

slowest quantum vs fastest classical

The one chart that tells the story

Each bar spans the 95% bootstrap confidence interval for ROC-AUC. The dashed line marks 0.5, the score of a coin flip. Classical bars sit clearly to the right of chance; every quantum bar crosses it.

Explore

Findings

The full results table, confidence intervals, and the five honest takeaways.

Methodology

Dataset, preprocessing, feature maps, quantum circuits, and how uncertainty is estimated.

Architecture

How the package is laid out and how data flows from raw CSV to results and figures.

Live demo

Enter ICU values and get an instant risk estimate from a simplified classical model.

Reproducibility

One command rebuilds every number, figure, and table from a fixed seed.

Limitations

What this benchmark does not claim, and where the quantum result could change.

Not a clinical tool

This is a research and portfolio project on synthetic or de-identified competition data. Nothing here is validated for clinical use.