Part 1: Preprocessing MIMIC-IV for Readmission Prediction

Author(s): Marie Humbert-Droz, PhD

Originally published on Towards AI.

Kicking off a hands-on series on building explainable AI models for healthcareImage generated with ChatGPT

In my last series, we tackled a critical question: How do we detect hallucinations in large language models built for clinical use?

But hallucinations aren’t the only trust problem in healthcare AI. While predictive models don’t hallucinate, they have other issues: lack of transparency.

Machine learning models typically output a prediction — and based on training and validation data, we can estimate how accurate that prediction is overall. But what we often can’t see is how the model reached its conclusion

Which features mattered most?Why this patient?Why now?

In healthcare, that lack of visibility isn’t just inconvenient — it’s a serious risk. That’s the black box problem.

In this new series, we’ll tackle that challenge directly. We’ll build an explainable machine learning pipeline for hospital readmission prediction — using real-world MIMIC-IV data — and show how methods like SHAP and LIME can bring transparency to model decisions.

In healthcare, a correct prediction isn’t enough. Clinicians need to understand why a model made a recommendation before they’ll act on it. That’s where explainable AI (XAI) comes in:

Clinical Trust and Adoption: Clinicians need to understand model reasoning before acting on it.Regulatory Compliance: The FDA now recommends transparency and audit trails for AI/ML medical devices (FDA… Read the full blog for free on Medium.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI