The objective of the challenge is to build a predictive model that accurately classifies the presence of chronic atrial fibrillation or other arrhythmias in a sleep session, given the time series of RR intervals.

RR intervals can be generated from wearables - e.g. from ECG sensors found on chest straps. A solution that warns a person of heart issues using data from normal sleep sessions could alert them to an issue and allow for early treatment.

The data has been split into 349 readings for the training set and 166 for the test set.

The data has been stored in csv format. Each file consists of RR intervals from one night’s sleep recorded in a sleep laboratory during a polysomnography test (sleep study).

• RR intervals have been extracted from 200Hz ECG data. Note: sometimes, the ECG signal is unreliable leading to missing or inaccurate data. You should carefully consider which RR interval values are reliable.
• The RR-interval time series covers the full duration of the session including:
• The periods the participant was awake before and after the sleep session
• Periods when the participant was awake during the sleep session
• The zipped ‘train.zip’ folder includes sleep study files labelled ‘rr_xxxxxx.csv’ (where ‘xxxxxx’ is a unique hash used to identify each sleep study), containing the RR interval values in milliseconds with corresponding relative timestamps in milliseconds. You can assume the sleep study starts at 0ms for each file.
• he target ‘training_target.csv’ file, contains the expertly hand-labelled arrhythmia class labels in the ‘target’ column (see table below). The ‘ID’ column matches the labelling of the training files containing RR intervals.

📷

• The zipped ‘test.zip’ folder contains the sleep study files that you will apply your model to. This dataset contains the same RR interval type files as the training data. There is no arrhythmia target file - this is what you are predicting.

Variable definitions

• ms: the timestamp, in milliseconds, since the start of the data collection
• rr_interval: the time interval, in milliseconds, between successive R-waves of the QRS signal
• ID: unique participant’s sleep session identifier file name starting with rr_
• target: arrhythmia class label (0=healthy, 1=afib, 2=other)

Files available for download:

• train.zip - contains .csv files of RR intervals . This is the dataset that you will use to train your model.
• training_target.csv - contains the target class for each training file
• test.zip - resembles ’train.zip’ except there is no matching target file. This is the dataset on which you will apply your model to.
• sample_submission.csv - shows the submission format for this competition, with the ‘ID’ column mirroring the RR interval filename in test.zip and the ‘target’ column contains your arrhythmia classification (0=healthy, 1=afib or 2=other) per test file. The order of the rows does not matter, but the names of the ‘ID’ must be correct.

Reference:

Goldberger, A., Amaral, L., Glass, L., Hausdorff, J., Ivanov, P. C., Mark, R., ... & Stanley, H. E. (2000). PhysioBank, PhysioToolkit, and PhysioNet: Components of a new research resource for complex physiologic signals. Circulation [Online]. 101 (23), pp. e215–e220.