Timeseries
Here we describe our continuous timeseries data: Heart Rate, Activity Count, Activity Type, Steps, Skin Conductance, and Skin Temperature. These streams are sampled at different granularities and have specific data characteristics.
The following table summarizes each metric:
| Metric | Granularity | Units | Range | Description |
|---|---|---|---|---|
| HEART_RATE | 10s | bpm | 40–220 | Continuous Heart Rate. Includes a quality indicator. |
| ACTIVITY_COUNT | 1s | – | 0–63000 | A unidimensional scale of movement. Higher values indicate greater movement. |
| ACTIVITY_TYPE | until value changes | – | N.A. | Activity Type (UNSPECIFIED, OTHER, WALK, RUN, CYCLE, REST). Includes a quality indicator. |
| STEPS | when steps are detected | steps | 12–N | Number of steps detected. |
| SKIN_CONDUCTANCE_LEVEL | 5s | µS | 0-100 | Continuous skin conductance. |
| SKIN_TEMPERATURE | 5s | ºC | 20-50 | Skin temperature. |
| REACTIVITY_MONITOR | 1 minute | – | 0-100 | A score of how reactive the user is to stressors. |
| MOTION | 1 minute | – | 0-100 | A score of how much the user has moved in the last minute. |
Note
Activity Type: data requires forward filling to obtain values at every second. You only obtain a new value when the quality indicator or the activity type values change. - Steps: sent in batches when at least 12 new steps are recorded since the previous batch.
- Skin Temperature: skin temperature is not the same as core body temperature.
Details about these metrics
- Heart Rate Quality Indicator
- Ranges from 1 to 4 (4 = best). Values of 1–4 are valid; data with quality 0 are considered not to be reliable and are therefore should be discarded. Motion can significantly affect HR quality.
-
Additional information about activity count.
- Although activity count is an unidimensional measure of movement, approximate average guidelines (per minute) are:
- Fake data (still on top of a table):
< 50 - Sleeping/meditating:
< 250(excluding moments wher user moves) - Sitting, low arm movement:
250-1500 - Sitting , working on the computer:
1500-4000 - Physical Activity:
> 4000 - Cycling:
4000-7000 - Walking:
6000-13000 - Running:
>10000
- Fake data (still on top of a table):
Note
These are average values and can significantly vary by user.
- Although activity count is an unidimensional measure of movement, approximate average guidelines (per minute) are:
-
Activity Type Quality Indicator:
- Ranges from 1 to 4 (4 = best). More stable activity types over time tend to have higher quality values. Data with quality 0 is considered not reliable and should be discarded.
- Activity Type Value mapping:
- OTHER represents activities other than walking, running, cycling, or resting.
- UNSPECIFIED means missing or undetectable data.
- Data with quality 0 is considered not reliable and should be discarded.
- Activity Type Preprocessing:
- A new activity type value is sent only when either the activity type or its quality indicator changes. To get continuous data at 1-second resolution, forward-fill the last known value for missing seconds.
- Interbeat Interval (IBI) Quality Indicator:
- Ranges from 1 to 4 (4 = best). IBI values are only calculated when consecutive heartbeats can be reliably detected from high-quality PPG signals. Values outside the 100-2000ms range are filtered out as physiologically implausible. When multiple IBIs occur at the same timestamp, their values are averaged and the lowest quality indicator is used.
- IBI data is event-based and irregular, as it depends on the detection of consecutive heartbeats. The availability of IBI data is highly dependent on motion levels, with more data available during periods of rest or low activity.