# HAR70+: Thigh and Lower-Back Accelerometry of 18 Adults Aged 70 to 95 with Observer-Annotated Daily Activities (NTNU)

## Facts

- Provider: Norwegian University of Science and Technology (Ustad, Logacjov, Trollebø, Thingstad, Vereijken, Bach, Mork); harmonized by Josef Heidler
- License: CC-BY-4.0 (https://archive.ics.uci.edu/dataset/780)
- Size: 20 files
- Format: Parquet (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/datasets/josefheidler/har_older-adults_2023-harth
- Access: public, readable without an account
- Category: Physiology
- Source: https://doi.org/10.3390/s23010130
- Page: https://scigantic.com/archive/f8dc00d9-5f7c-5bcc-81fc-6e294f736dd4
- Keywords: Physiology, Parquet, Hugging Face

## About

HAR70+ (Human Activity Recognition 70+) from NTNU in Trondheim: raw triaxial accelerometry of 18 older adults aged 70 to 95, five of whom used walking aids, recruited from a rehabilitation clinic. Each wore two Axivity AX3 accelerometers (plus or minus 8 g, 50 Hz) on the right thigh, about 10 cm above the kneecap, and on the lower back near the third lumbar vertebra during a roughly 40-minute semi-structured free-living session, while a trained observer logged activity in real time. Seven activity types are labelled: walking, shuffling, stair ascent, stair descent, standing, sitting and lying. The release has about 2.26 million samples and 12.6 labelled hours, all samples labelled. It is the training set for activity type classifiers for older adults and frail populations, gait and mobility research, and transfer from the younger-adult HARTH dataset. This Hugging Face copy harmonizes the NTNU release into one Parquet file per participant under harmonized/, with back and thigh x, y, z acceleration in g and the activity label for every sample; load with Hugging Face datasets or pandas. The UCI Machine Learning Repository entry for HAR70+ (DOI 10.24432/C5CW3D) states the CC BY 4.0 license. Cite Ustad et al., Sensors 23(1):130, 2023.

## Load the data

```python
from datasets import load_dataset
ds = load_dataset("josefheidler/har_older-adults_2023-harth")
print(ds)
```

## Data structure

File types in sampled files: Parquet

Example file: `harmonized/agile-anteater.parquet`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/f8dc00d9-5f7c-5bcc-81fc-6e294f736dd4
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/f8dc00d9-5f7c-5bcc-81fc-6e294f736dd4
- Over MCP: `get_schema_card` with id `f8dc00d9-5f7c-5bcc-81fc-6e294f736dd4` at https://api.scigantic.com/mcp

## Related topics

- [Parquet datasets](https://scigantic.com/topics/parquet)
- [Physiology datasets](https://scigantic.com/topics/physiology)
