Skip to content

MBA (MIT-BIH Supraventricular Arrhythmia, ECG)

What is this data? Two-lead ECG recordings from PhysioNet's MIT-BIH database with cardiologist beat annotations; anomalies are the non-normal beats (arrhythmia). The signal is near-periodic, which makes it the classic showcase for subsequence methods — and a frequent appendix dataset in TSAD papers (TranAD, among others).

  • Source: processed split shipped in https://github.com/imperial-qore/TranAD (data/MBA); underlying recordings: PhysioNet MIT-BIH
  • Contents: 2 channels (ECG1/ECG2), ~7.7k train / 7.7k test steps; labels derived from beat annotations (±20 samples around each non-'N' beat — the TranAD convention)
  • License: TranAD repository BSD-3-Clause; underlying PhysioNet data ODC-BY (open)
  • Download: tsad-forge download mba
  • Loader: load_mba()

Known flaws / caveats

  • High anomaly rate (~34%) — far from the rare-anomaly regime; precision-oriented metrics behave differently here. Read alongside low-rate datasets.
  • Labels are beat-centered fixed windows (±20 samples), not exact episode boundaries.
  • This is the small TranAD excerpt, not the full MIT-BIH database.

Quick EDA (from local data)

Sample test channels with labeled anomalies shaded, plus the event-length distribution. Regenerate with tsad-forge viz after downloading the data.