Synthetic Data Guided Feature Selection for Robust Activity Recognition in Older Adults

TL;DR

The paper tackles robust human activity recognition (HAR) for older adults during hip fracture rehabilitation, where standard trackers struggle with slow and variable gait patterns. It introduces a synthetic data guided feature intervention model (FIM) that uses dynamic time warping barycentre averaging (DBA) to generate generalized gait representations, guiding feature selection before applying to real data. Using 24 participants aged 80+, two accelerometer placements, and leave-one-subject-out cross-validation, the FIM outperformed a real-data-only baseline, with notable improvements in transfer detection. These results demonstrate improved generalization across individuals and lay groundwork for validation in hip fracture patient populations, addressing a critical need for continuous activity monitoring in geriatric rehabilitation.

Abstract

Physical activity during hip fracture rehabilitation is essential for mitigating long-term functional decline in geriatric patients. However, it is rarely quantified in clinical practice. Existing continuous monitoring systems with commercially available wearable activity trackers are typically developed in middle-aged adults and therefore perform unreliably in older adults with slower and more variable gait patterns. This study aimed to develop a robust human activity recognition (HAR) system to improve continuous physical activity recognition in the context of hip fracture rehabilitation. 24 healthy older adults aged over 80 years were included to perform activities of daily living (walking, standing, sitting, lying down, and postural transfers) under simulated free-living conditions for 75 minutes while wearing two accelerometers positioned on the lower back and anterior upper thigh. Model robustness was evaluated using leave-one-subject-out cross-validation. The synthetic data demonstrated potential to improve generalization across participants. The resulting feature intervention model (FIM), aided by synthetic data guidance, achieved reliable activity recognition with mean F1-scores of 0.896 for walking, 0.927 for standing, 0.997 for sitting, 0.937 for lying down, and 0.816 for postural transfers. Compared with a control condition model without synthetic data, the FIM significantly improved the postural transfer detection, i.e., an activity class of high clinical relevance that is often overlooked in existing HAR literature. In conclusion, these preliminary results demonstrate the feasibility of robust activity recognition in older adults. Further validation in hip fracture patient populations is required to assess the clinical utility of the proposed monitoring system.

Figures (7)

• Figure 1: Schematic of activity trackers’ placements on the upper thigh and lower back.
• Figure 2: General workflow of three proposed human activity recognition models. The process blocks involving the direct usage of synthetic data or derived features are colored in green.
• Figure 3: General protocol for the generation of synthetic data using dynamic time warping barycentre averaging (DBA), on both MOX and APDM sensory data.
• Figure 4: Overview of the multistep feature selection (FS) pipeline. The pipeline was applied to the real data to build the control condition model, and it was applied to the synthetic data to build the complete data intervention model and the feature intervention model. Five FS algorithms were selected in this work, including Relief-F urbanowicz2018relief, maximum relevance minimum redundancy (MRMR) ding2011minimum, interaction-curvature tests embedded into a decision tree (ICT) loh2002regression, out-of-bag feature importance by permutation embedded into a random forest (OOB-I) breiman2001random, and regularization embedded into a linear discriminant (LD-R) guo2007regularized.
• Figure 5: Overview of feature selection results using the heterogeneous feature selection ensemble (HFSE) across the 10 subsamples of the real and synthetic data. X and Z denote the acceleration axes. UT, LB, and RMS are abbreviations for upper thigh, lower back, and root-mean-square, respectively.