Research Appraisaldiagnostic

Recognition of everyday activities using experiment data from wearable sensors: a deep learning-based framework

Scientific reportsGalanza, FrameworkWilliam Son, Schmidt, Steven M, Fristedt, Sofi et al.24 July 2026DOI

Clinical Snapshot

35CEBM
Evidence: Weakdiagnostic

PICO Framework

P — PopulationOlder adults (aged 65 and over, including aged 80 and over) participating in a small-scale home laboratory experiment
I — InterventionWearable inertial sensor systems (one, two, or five sensors at various body placements) combined with deep learning classification models to recognise everyday activities
C — ComparatorFive competing deep learning model configurations varying in sensor count (one, two, or five sensors) and sensor placement sites
O — OutcomesClassification accuracy of 14 everyday activities (primary); identification of optimal sensor placement and minimum sensor count for reliable activity recognition

Bottom Line

This proof-of-concept study from Lund University proposes a two-sensor wearable system (pelvis and right hand) using deep learning to classify everyday activities in older adults, reporting 89.3% accuracy across 12 activities. While the clinical motivation is sound — passive monitoring of functional activity to detect early health deterioration in older adults is a genuine and growing need — the study has substantial methodological limitations that preclude clinical translation at this stage. The sample size is undisclosed but acknowledged as small, no confidence intervals are reported, and the controlled laboratory setting bears little resemblance to real-world home environments. The absence of per-activity sensitivity and specificity data is a critical gap; overall accuracy can mask dangerous misclassification of clinically important activities such as falls or prolonged inactivity. Ground truth labelling methods and sensor specifications are not described. External validation in a representative community-dwelling older adult population is essential before this approach can be considered for clinical or aged care deployment. For Australian clinicians and aged care providers, this study represents an interesting early-stage development but should not influence current practice. Future work must include adequate sample sizes, real-world validation, and evaluation of patient-centred outcomes including acceptability and downstream health impact.

Evidence: Weak

Key Findings

  • P Value: Not reported

  • Effect Size: Two-sensor model (pelvis and right hand): 89.3% accuracy for 12 activities; Five-sensor model: 88.2% accuracy for 14 activities; One-sensor model: low recognition performance (specific figures not reported in abstract)

  • Primary Outcome: Classification accuracy of everyday activities using deep learning models with varying wearable sensor configurations

  • Nnt Or Sensitivity: Per-activity sensitivity and specificity not reported; overall accuracy only. Analogous to a diagnostic accuracy study without reporting sensitivity/specificity breakdown — a critical omission for clinical translation

  • Confidence Interval: Not reported

Clinical Application

A two-sensor wearable system (pelvis and right hand) is conceptually feasible for older adults if devices are lightweight, comfortable, and require minimal user interaction. However, adherence, charging requirements, skin integrity concerns, and cognitive capacity to manage devices in this population have not been evaluated. Integration with existing clinical workflows and electronic health records would require substantial additional development. Australia's rapidly ageing population and the National Aged Care Strategy create genuine demand for scalable remote monitoring tools. The My Aged Care framework and RACGP's Silver Book guidelines emphasise functional assessment in older adults, which this technology could theoretically support. However, TGA regulatory approval as a Software as a Medical Device (SaMD) would be required before clinical use. PBS subsidy pathways for wearable monitoring systems are currently limited. The Australian Digital Health Agency's interoperability standards would need to be met for integration with My Health Record. Aboriginal and Torres Strait Islander older adults and those from CALD backgrounds are underrepresented in technology development studies of this type, raising equity concerns. This study does not yet meet the evidentiary threshold for RACGP or AHPRA-endorsed clinical recommendation. Community-dwelling older adults requiring passive health monitoring for early detection of functional decline; potentially applicable in aged care facilities, post-hospitalisation monitoring, and chronic disease management programs. Current evidence supports proof-of-concept only — not yet ready for clinical deployment.

Abstract

Tracking everyday activities is vital for detecting changes in older adults' health, allowing timely support to promote well-being. Wearable sensors and deep learning provide continuous monitoring, making them a supportive tool in detecting such changes. However, a more refined method is needed to recognise precise activities with a minimal set of sensors. This study aimed to develop a method to recognise everyday activities among older adults by utilising wearable sensors and a deep learning model. This is a small-scale home lab experiment to develop a method to recognise 14 everyday activities. We compared five models that recognised everyday activities with different sensor signal counts and accuracy. Our results showed that sensor placement is important. Based on the results, we proposed a two-sensor method (pelvis and right hand) to collect and correctly recognise everyday activities among older adults. This model, which utilises two sensors, classified 12 activities with an accuracy of 89.3%. Another model recognised all 14 activities with a lower accuracy of 88.2% using five sensors. We also explored a one-sensor approach, which showed low recognition performance and struggled to distinguish activity variability. The two-sensor-based system will allow for large-scale data collection on everyday activities of older adults.

References

  1. 1.Galanza, W. S., Schmidt, S. M., Fristedt, S., & Malesevic, N. (2026). Recognition of everyday activities using experiment data from wearable sensors: a deep learning-based framework. Scientific Reports. https://doi.org/10.1159/000530900
Share:XLinkedIn

This content is for educational purposes for healthcare professionals only and does not constitute clinical advice. Clinical decisions should be based on individual patient assessment, current guidelines, and appropriate specialist consultation. Editorial Standards · Privacy Policy · Terms of Service