Characterizing the Performance Gap in Human Activity Recognition for Older Adults

Hossein Khayami, Sungjin Hwang, Eshed Ohn-Bar, David E. Conroy, Amanda Lazar, Eun Kyoung Choe, Hernisa Kacorri

Summary

Uses MyMove — authors' annotated free-living older-adult wrist-accelerometer HAR dataset (mean age 71) — to test architectures/training regimes under LOSO and cross-dataset transfer. Gains on younger-adult benchmarks do not transfer equally; the gap persists or widens. Frozen self-supervised features pretrained on age-diverse UK Biobank accelerometry substantially narrow the gap (small cost on younger adults); disparities remain.