New paper for the 31st European Conference on Information Systems: „Best of both worlds: Combining predictive power with interpretable and explainable results for patient pathway prediction”
We as a chair are happy to announce that our latest paper „Best of both worlds: Combining predictive power with interpretable and explainable results for patient pathway prediction” has been accepted for presentation at the 31st European Conference on Information Systems.
Sandra Zilker, Sven Weinzierl, Patrick Zschech, Mathias Kraus and Martin Matzner, the authors of this paper, present a novel artifact called HIXPred for patient pathway prediction. HIXPred combines predictive power with interpretability where possible, and explainability where necessary.
They developed a deep learning model consisting of a BI-LSTM layer for sequential features and a simple feed-forward layer for static features. Explainability is created for the sequential features using the post-hoc XAI method SHAP, and interpretation for the static features refers to coefficients directly extracted from the feed-forward layer.
They applied HIXPred to a real-life use case for evaluation and demonstration purposes. Evaluation and demonstration results show that HIXPred can provide high predictive performance while ensuring sufficient interpretability and explainability. Furthermore, interviews with medical experts confirm the usefulness of our artifact.
The paper can be found here.
We are looking forward to exciting and fruitful discussions in Kristiansand.