geronimobasso/apple-watch-ecg-export-data

Dataset Card for Apple Watch Lead I ECG (single subject)

This dataset card aims to be a base template for new datasets. It has been generated using this raw template.

Dataset Details

Dataset Description

Ten 30-second single-lead (Lead I) electrocardiogram recordings captured by an Apple Watch over roughly three years (2023-2025). Each recording is a single column of microvolt samples at 512 Hz, distributed as CSV.

• Curated by: Geronimo Basso Sosa
• Language(s) (NLP): English
• License: MIT

Uses

Direct Use

• Consumer-wearable ECG signal processing demos.
• Teaching and educational examples for single-lead ECG handling.
• Time-series preprocessing, denoising, and feature extraction experiments.

Out-of-Scope Use

• Clinical diagnosis or medical decision-making.
• Any use that treats this single-subject dataset as population-representative.
• Building or validating high-stakes health models without broader, diverse clinical data.

Dataset Structure

• Number of recordings: 10 CSV files.
• Recording duration: 30 seconds each.
• Sampling rate: 512 Hz.
• Samples per file: 15,359.
• Total samples: 153,590.
• Lead: Lead I.
• Unit: microvolts (uV).
• File pattern: ecg-*.csv.
• Filenames are opaque sequential IDs assigned in chronological order from oldest to newest (ecg-1.csv through ecg-10.csv).
• Column schema (all files):

Column	Type	Description
lead_i_uv	float	Lead I ECG amplitude in microvolts sampled at 512 Hz

• Rows do not carry explicit timestamps. Time can be reconstructed from sample index and sampling rate.

Dataset Creation

Curation Rationale

The dataset was assembled as a compact, well-documented reference for wearable ECG signal processing, teaching, and demos.

Source Data

The recordings originate from the Apple Watch ECG app and were exported from Apple Health data.

Data Collection and Processing

• ECG sessions are fixed 30-second recordings produced by the Apple Watch ECG app.
• Data is provided as one CSV per recording with a single numeric column.
• Two Apple Watch hardware identifiers appear in the source export (Watch6,7 and Watch7,8).

Who are the source data producers?

• A single human subject produced all recordings.

Annotations [optional]

Annotation process

No additional manual annotation process is documented.

Who are the annotators?

No annotators. The dataset consists of direct device signal exports.

Personal and Sensitive Information

This dataset contains health-related biosignal data (ECG), which is sensitive personal information. The dataset is single-subject and therefore carries re-identification risk. Users should avoid attempting re-identification and should not use the dataset in ways that could harm privacy.

Bias, Risks, and Limitations

• Single-subject dataset with very limited sample size (10 recordings).
• Data comes from a consumer wearable single-lead setup, not a clinical 12-lead ECG workflow.
• Potential device- and hardware-specific noise characteristics.
• Not suitable for broad generalization across populations.

Recommendations

Users should be made aware of the risks, biases and limitations of the dataset. More information needed for further recommendations.

Glossary [optional]

• Lead I: Voltage difference between left arm (positive) and right arm (negative).
• Sampling rate (512 Hz): 512 samples per second.
• uV: Microvolts, the unit used for ECG amplitude.

Dataset Card Authors [optional]

Geronimo Basso Sosa

Dataset Card Contact

[More Information Needed]

Hardware-specific noise characteristics differ slightly between the two models — visible in overview.png (the 2025-12 trace is the noisiest).