YAIB
A holistic framework for standardizing clinical prediction model experiments on ICU and EHR data.
Standardize clinical machine learning experiments across multiple ICU datasets and prediction tasks.
YAIB (Yet Another ICU Benchmark) is a Python framework designed to standardize clinical machine learning experiments using Intensive Care Unit and Electronic Health Record data. It suits researchers and data scientists working on predictive modeling across multi-center datasets like MIMIC, eICU, HiRID, and AUMCdb.
What it is
The framework provides out-of-the-box support for major ICU databases and five standard clinical tasks, including ICU mortality, acute kidney injury, sepsis, kidney function, and length of stay. It operates on preprocessed parquet files and integrates with PyTorch and PyTorch Lightning for deep learning workflows.
Who it's for
It suits machine learning researchers and clinical data scientists who need a unified pipeline to benchmark predictive models across different international ICU databases. It does not suit clinicians without programming experience or those looking for an out-of-the-box clinical decision support software.
Key features
Multi-Dataset Support
Out-of-the-box compatibility with MIMIC-III, MIMIC-IV, eICU-CRD, HiRID, and AUMCdb databases.
Standard Clinical Tasks
Includes five default tasks covering binary classification and regression endpoints like mortality and sepsis.
Parquet-Based Pipeline
Designed to operate efficiently on preprocessed parquet files with built-in cache generation and loading.
PyTorch Integration
Built on PyTorch 2.0+ and Lightning 2.0+ for scalable deep learning and model experimentation.
Up and running in 3 steps
Our honest verdict
What's great
- Standardizes evaluation across disparate international ICU databases
- Includes pre-configured demo datasets for immediate pipeline testing
- Supports both classical machine learning models and deep learning
- Provides modular companion repositories for cohorts, models, and preprocessing
Where it falls short
- Requires technical skill with Python, CLI tools, and package managers like uv
- Depends on gaining credentialed access to external PhysioNet databases for full usage
- Running heavy multi-center deep learning experiments requires significant GPU compute resources
- Data preprocessing pipelines require substantial manual setup before ingestion
Pricing & license
Free under the MIT license, with no paid hosted options provided. The only cost is running it yourself and acquiring the necessary hardware compute for training models.
Open remote AI-training roles
870 open remote AI-training roles, updated November 14, 2026; most pay $20–$170 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
These reviewed AI tools do the same job instantly — from our scored directory.
Use it, or have us set it up
FAQ
What datasets are supported out of the box?
YAIB supports MIMIC-III, MIMIC-IV, eICU-CRD, HiRID, and AUMCdb datasets.
Can I add custom datasets and tasks?
Yes, the framework is designed to let you provide custom datasets, cohorts, prediction tasks, and endpoints.
Do I need access to real ICU data to test YAIB?
No, the repository includes processed demo datasets from eICU and MIMIC so you can test training commands immediately.
