Train SizeTest SizeLength Number of ClassesNumber of DimensionsType
478785 90315178 51EEG
Data Source: Link Here
Donated By: Paul Rabich
Description:

Sleep contains 153 whole-night sleeping Electroencephalography (EEG) recordings taken from physionet (Link Here) presented in [1] and formatted into a classification problem in [2]. The data is collected from 82 healthy subjects. The 1-lead EEG signal is sampled at 100 Hz.

The series were segmented into non overlapping sub series, each of which forms a case. Each case is labelled with one of the five sleeping patterns/stages: Wake (W), Nonrapid eye movement (N1, N2, N3) and Rapid Eye Movement (REM). The classes are not balanced and there are differences in the class distribution for the train and test. Train/test proportions in brackets after name.

  1. Wake (13.84% in train and 4.05% in test)
  2. Non rapid eye movement type 1 (6.56% in train and 10.98% in test)
  3. Non rapid eye movement type 2 (42.73% in train and 48.93% in test)
  4. Non rapid eye movement type 3 (15.81% in train and 18.82% in test)
  5. Rapid Eye Movement (21.06% in train and 18.82% in test)

The data were split in [2] into 371,055 train cases, 107,730 validation and 90,315 test. (Link Here) We have added the validation set to the end of the train file to ease reproduction if a validation set is needed. For reference, a MiniRocket classifier gets accuracy of 92.7% on the default test data.

[1] B Kemp et al. Analysis of a sleep-dependent neuronal feedback loop: the slow-wave microcontinuity of the EEG. IEEE-BME 47(9):1185-1194 (2000).

[2] Zhang et al. Self-Supervised Contrastive Pre-Training For Time Series via Time-Frequency Consistency, NeurIPS 2022. (Link Here)

Download this dataset
Dataset Image

Best Algorithm:
Best Accuracy:
Second Link: Link Here