Skip to main content

Table 6 PAMAP2 and MHEALTH to PD cross-testing binary accuracy and F1 Scores with Domain Adaptation

From: Deep neural networks for wearable sensor-based activity recognition in Parkinson’s disease: investigating generalizability and model complexity

 

Base Model

O’Halloran [13]

Wan [14]

Burns [15]

Chen/Jordao [10, 16]

Kalouris [17]

Yatbaz [18]

Rueda [19]

\(\begin{array}{l}\hbox {Source: P2 Aug} \\ \hbox {Target: PD Reorient}\end{array}\)

\(\begin{array}{l}40.01\\ (6.06)\end{array}\)

\(\begin{array}{l}62.45\\ (21.43)\end{array}\)

\(\begin{array}{l}64.83\\ (43.65)\end{array}\)

\(\begin{array}{l}66.67\\ (44.62)\end{array}\)

\(\begin{array}{l}70.19\\ (45.18)\end{array}\)

\(\begin{array}{l}73.02\\ (49.79)\end{array}\)

\(\begin{array}{l}69.79\\ (44.08)\end{array}\)

\(\begin{array}{l}67.02\\ (0.04)\end{array}\)

\(\begin{array}{l}\hbox {Source: MH Aug} \\ \hbox {Target: PD Reorient}\end{array}\)

\(\begin{array}{l}61.89\\ (19.99)\end{array}\)

\(\begin{array}{l}69.10\\ (37.49)\end{array}\)

\(\begin{array}{l}63.35\\ (24.13)\end{array}\)

\(\begin{array}{l}67.49\\ (20.16)\end{array}\)

\(\begin{array}{l}57.00\\ (19.58)\end{array}\)

\(\begin{array}{l}62.80\\ (29.35)\end{array}\)

\(\begin{array}{l}66.23\\ (34.01)\end{array}\)

\(\begin{array}{l}68.09\\ (3.97)\end{array}\)

  1. The average values are shown, with F1 scores presented in parentheses. The notation for data augmentation applies to both the MHEALTH and PAMAP2 datasets. ‘PD Reorient’ indicates PD data with sensor axes reoriented to match the source data