Table 2 Selected datasets and their main features.

From: A benchmark for domain adaptation and generalization in smartphone-based human activity recognition

Acronym

Dataset Name

Smartphone Position

Accelerometer Metricb

Sampling Rate

Number of Users/Activ

KH

KU-HAR23

waist bag

m/s2

100 Hz

90/18

MS

MotionSense24

pocket

G

50 Hz

24/6

RW

RealWorld3

thigh/waista

m/s2

50 Hz

15/8

UCI

UCI-HAR25

waist bag

G

50 Hz

30/5

WDM

WISDM2

pocket

m/s2

20 Hz

51/18

  1. aThis dataset employs multiple smartphones placed in different locations during data collection (e.g., thigh, waist, shin, head, etc.). We used data collected by sensors placed at thigh (RW-Thigh) and waist (RW-Waist), which should be equivalent to a pocket and a waist bag.
  2. bAll datasets record gyroscope samples in rad/s.
Back to article page