DeltaSHAP: Explaining Prediction Evolutions in Online Patient Monitoring with Shapley Values (ICMLW-AIW 2025)

DeltaSHAP: Explaining Prediction Evolutions in Online Patient Monitoring with Shapley Values
Changhun Kim*, Yechan Mun*, Sangchul Hahn, Eunho Yang (*: equal contribution)
ICML Workshop on Actionable Interpretability, 2025

Introduction

This repository contains the official PyTorch implementation of DeltaSHAP: Explaining Prediction Evolutions in Online Patient Monitoring with Shapley Values. DeltaSHAP is a fast, faithful, and model-agnostic explanation method that attributes changes in real-time risk predictions to newly observed clinical features using temporal Shapley values. Built upon WinIT—with sincere thanks to the original authors.

Environmental Setup

conda create -n deltashap python=3.10.9
conda activate deltashap
pip install -r requirements.txt

Reproducing Experiments

bash scripts/run.sh

Contact

If you have any questions or comments, feel free to contact us via [email protected].

Citation

@inproceedings{kim2025deltashap,
  title={{DeltaSHAP: Explaining Prediction Evolutions in Online Patient Monitoring with Shapley Values}},
  author={Kim, Changhun and Mun, Yechan and Hahn, Sangchul and Yang, Eunho},
  booktitle={ICML Workshop on Actionable Interpretability},
  year={2025}
}