HiRID-ICU-Benchmark -- A Comprehensive Machine Learning Benchmark on\n High-resolution ICU Data

Abstract

The recent success of machine learning methods applied to time series\ncollected from Intensive Care Units (ICU) exposes the lack of standardized\nmachine learning benchmarks for developing and comparing such methods. While\nraw datasets, such as MIMIC-IV or eICU, can be freely accessed on Physionet,\nthe choice of tasks and pre-processing is often chosen ad-hoc for each\npublication, limiting comparability across publications. In this work, we aim\nto improve this situation by providing a benchmark covering a large spectrum of\nICU-related tasks. Using the HiRID dataset, we define multiple clinically\nrelevant tasks in collaboration with clinicians. In addition, we provide a\nreproducible end-to-end pipeline to construct both data and labels. Finally, we\nprovide an in-depth analysis of current state-of-the-art sequence modeling\nmethods, highlighting some limitations of deep learning approaches for this\ntype of data. With this benchmark, we hope to give the research community the\npossibility of a fair comparison of their work.\n