CMIDSS Dataset

Overview

This dataset comes from the Child Mind Institute - Detect Sleep States (CMIDSS) Kaggle competition. The dataset comprises 300 subjects with over 500 multi-day recordings of wrist-worn accelerometer data annotated with two event types: onset, the beginning of sleep, and wakeup, the end of sleep. While the original data contains 3-axis accelerometer data, this dataset only contains the euclidean norm minus one (ENMO) and z-angle reported every 5 seconds.

More info available on Kaggle competition page

Funding

The data was provided by the Healthy Brain Network, a landmark mental health study based in New York City that will help children around the world. In the Healthy Brain Network, families, community leaders, and supporters are partnering with the Child Mind Institute to unlock the secrets of the developing brain. In addition to generous support provided by the Kaggle team, financial support has been provided by the Stavros Niarchos Foundation (SNF) as part of its Global Health Initiative (GHI) through the SNF Global Center for Child and Adolescent Mental Health at the Child Mind Institute.

Licensing

The CMIDSS dataset is available for non-commercial use under Attribution-NonCommercial-ShareAlike 4.0 International.

Task integration

Use CMIDSS for sleep detection. Its onset/wakeup annotations do not provide sleep-stage or apnea labels. See the detection label policy for the derived target used by the experimental recipe.

Installation

The dataset adapter has two download paths. download_raw_dataset() retrieves the competition data through the Kaggle client and converts subject recordings to HDF5. Configure the Kaggle client and obtain competition access before running:

Python example

import sleepkit as sk

dataset = sk.DatasetFactory.get("cmidss")(path="./datasets/cmidss")
dataset.download_raw_dataset()

dataset.download() instead retrieves the prepared archive from the configured Ambiq S3 dataset bucket. It does not use the Kaggle client or an NSRR token. Access to that archive is separate from competition access.