Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 45 of the 45 references it cites were found in a registry or an index; 7 of the 59 sources its claims cite were checked for support.

Towards User-Generalizable Wearable-Sensor-Based Human Activity Recognition: A Multi-Task Contrastive Learning Approach

Open access
Guo, Pengyu, Nakayama, Masaya · 2025 · Sensors

Abstract

Human Activity Recognition (HAR) using wearable sensors has shown great potential for personalized health management and ubiquitous computing. However, existing deep learning-based HAR models often suffer from poor user-level generalization, which limits their deployment in real-world scenarios. In this work, we propose a novel multi-task contrastive learning framework that jointly optimizes activity classification and supervised contrastive objectives to enhance generalization across unseen users. By leveraging both activity and user labels to construct semantically meaningful contrastive pairs, our method improves representation learning while maintaining user-agnostic inference at test time. We evaluate the proposed framework on three public HAR datasets using cross-user splits, achieving comparable results to both supervised and self-supervised baselines. Extensive ablation studies further confirm the effectiveness of our design choices, including multi-task training and the integration of user-aware contrastive supervision. These results highlight the potential of our approach for building more generalizable and scalable HAR systems.

Overview

CoChat Summary

TL;DR

The paper proposes MultiSupConHAR, a single-stage multi-task framework that jointly optimizes activity classification and supervised contrastive learning using both activity and user labels to improve cross-user generalization in wearable-sensor HAR. It reports competitive or best macro-F1 on MobiAct and USC-HAD, with slightly lower performance than some supervised baselines on UCI-HAR.

Key findings

  • MultiSupConHAR achieved macro-F1 of 85.93% ± 1.23% on MobiAct and 76.84% ± 1.09% on USC-HAD, ranking first on those datasets, and 91.07% ± 2.09% on UCI-HAR, ranking second behind supervised CSSHAR (93.73% ± 1.02%) and CTBL (92.72% ± 1.48%).
  • In ablation under LOSO, the proposed multi-task model reached 86.01% on MobiAct, 93.16% on UCI-HAR, and 77.13% on USC-HAD, improving over supervised-only by 10.88%, 3.45%, and 5.78%, and over two-stage SupCon-only by 3.20%, 0.94%, and 9.99%, respectively.
  • Adding a projection head improved MobiAct F1 from 78.87% to 86.01%; batch size 256 and projection hidden dimension 256 performed best across datasets.
  • The optimal contrastive loss weight λ varied by dataset: 0.1 for MobiAct, 0.2 for USC-HAD, and 0.4 for UCI-HAR; λ below 0.5 generally performed better.

Methods

  • The framework uses a shared lightweight ResNet encoder with three stacked residual blocks, a classification head for cross-entropy activity classification, and a projection head used only during training for supervised contrastive loss.
  • Positive pairs are samples with the same activity label across different users; negative pairs are samples with different activity labels from the same user. The total loss is \(L_{total}=L_{CE}+\lambda L_{CL}\), optimized in a single-stage multi-task setup.
  • Experiments used three public datasets: MobiAct (66 participants; 11 ADLs used), UCI HAR (30 participants; 6 ADLs), and USC-HAD (14 participants; 12 ADLs). Data were downsampled to 50 Hz, split into 1 s windows with 50% overlap, and evaluated with user-disjoint 60/20/20 train/validation/test splits over three independent runs using macro-F1.

Limitations

  • The paper states that the framework relies on fully labeled data for both activity and user identities, and suggests semi-supervised, self-supervised, pseudo-labeling, or weak-supervision extensions to reduce annotation requirements.
  • The paper states that validation was limited to public datasets and that in-the-wild deployment and long-term evaluation are needed to test robustness to sensor placement and user lifestyle variation.
  • The paper states that the framework mainly targets cross-user generalization; broader cross-domain and cross-device transferability would require complementary strategies such as domain adaptation, meta-learning, or adversarial data generation. Evident from the results: UCI-HAR performance was below supervised CSSHAR and CTBL, and confusion remained for transitional or low-motion activities such as sitting/standing and elevator/standing.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.3390/s25226988

Cites

45 of 45 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Figure 1 Overview of the proposed Multi-Task Contrastive Learning Framework for User-Generalizable HAR.
Figure 1 Overview of the proposed Multi-Task Contrastive Learning Framework for User-Generalizable HAR.CC BY · PubMed Central
Figure 2 Model architecture. The encoder consists of residual blocks with convolutional layers, followed by global average pooling. The projector and classifier are built from MLP layers with non-linear activations and dropout.
Figure 2 Model architecture. The encoder consists of residual blocks with convolutional layers, followed by global average pooling. The projector and classifier are built from MLP layers with non-linear activations and dropout.CC BY · PubMed Central
Figure 3 Comparison of positive and negative pair construction strategies.
Figure 3 Comparison of positive and negative pair construction strategies.CC BY · PubMed Central
Figure 4 Effect of contrastive loss weight λ on model performance (F1 score) across different datasets.
Figure 4 Effect of contrastive loss weight λ on model performance (F1 score) across different datasets.CC BY · PubMed Central
Figure 5 Effect of batch size on F1 score across three datasets.
Figure 5 Effect of batch size on F1 score across three datasets.CC BY · PubMed Central
Figure 6 Effect of projection head on F1 score across three datasets.
Figure 6 Effect of projection head on F1 score across three datasets.CC BY · PubMed Central
Figure 7 Effect of hidden dimension size in the projection head on F1 score.
Figure 7 Effect of hidden dimension size in the projection head on F1 score.CC BY · PubMed Central
Figure 8 Comfusion Matrix.
Figure 8 Comfusion Matrix.CC BY · PubMed Central
Figure 9 t-SNE visualization of representations (UCI HAR).
Figure 9 t-SNE visualization of representations (UCI HAR).CC BY · PubMed Central
Figure 10 t-SNE visualization of representations (MobiAct).
Figure 10 t-SNE visualization of representations (MobiAct).CC BY · PubMed Central
Figure 11 t-SNE visualization of representations (USC-HAD).
Figure 11 t-SNE visualization of representations (USC-HAD).CC BY · PubMed Central

Verification

Claims the paper makes

4 Valid · 0 Not valid · 23 Not checked · 92 Own results and interpretations

61 more claims

Create a free account to see all 119 claims and to verify the 52 references still open.

See all claims — free account
  • It may state two separate claims as one. 45
  • The quote only partly states this claim. 1
  • Probably not a substantial claim. 17

Create a free account to see these claims.

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

3 of 3

Citations

Cites

Show all 36 references

Research works printed but not linked: 5

  1. Ravi, N., Dandekar, N., Mysore, P., and Littman, M. (2005, January 9–13). Activity Recognition from Accelerometer Data. Proceedings of the Seventeenth Conference on Innovative Applications of Artificial Intelligence, Pittsburgh, PA, USA.
  2. Guo, P., and Nakayama, M. (March, January 27). Transformer-Based Human Activity Recognition Using Wearable Sensors for Health Monitoring. Proceedings of the 9th International Conference on Biomedical Engineering and Applications (ICBEA), Seoul, Republic of Korea.
  3. Chen, T., Kornblith, S., Norouzi, M., and Hinton, G. (2020, January 13–18). A Simple Framework for Contrastive Learning of Visual Representations. Proceedings of the 37th International Conference on Machine Learning (ICML), Virtual Event.
  4. Khosla, P., Teterwak, P., Wang, C., Sarna, A., Tian, Y., Isola, P., Maschinot, A., Liu, C., and Krishnan, D. (2020, January 6–12). Supervised Contrastive Learning. Proceedings of the 34th International Conference on Neural Information Processing Systems (NeurIPS), Vancouver, BC, Canada.
  5. Anguita, D., Ghio, A., Oneto, L., Parra, X., and Reyes-Ortiz, J.L. (2013, January 24–26). A Public Domain Dataset for Human Activity Recognition Using Smartphones. Proceedings of the European Symposium on Artificial Neural Networks (ESANN), Bruges, Belgium.

Cited by

See all 6

Metadata

Authors
Guo, P., & Nakayama, M.
Affiliations
The University of Tokyo (JP)
Corresponding author
Nakayama, Masaya
Venue
Sensors
Publisher
MDPI AG
Published
2025-11-15
Volume
25
Issue
22
Pages
6988
ISSN
1424-8220 (ISSN-L)
OpenAlex
W4416392909
Cited by
6
References
36
Citations by year
2026: 6
FWCI
2.11
Type
journal article
Language
en
Open access
Locations
  • Sensors (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Topics
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Context-Aware Activity Recognition Systems
  • Physical Sciences › Computer Science › Artificial Intelligence › Domain Adaptation and Few-Shot Learning
  • Physical Sciences › Engineering › Biomedical Engineering › Gait Recognition and Analysis
Keywords
human activity recognition, wearable sensors, cross-user generalization, contrastive learning, multi-task learning, supervised contrastive learning, representation learning
MeSH terms
Machine Learning, Wearable Electronic Devices, Deep Learning, Algorithms, Humans, Human Activities

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-09-29, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-09-29 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (qwen3.8-omni-flash and gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Towards User-Generalizable Wearable-Sensor-Based Human Activity Recognition: A Multi-Task Contrastive Learning Approach. https://app.cochat.ai/papers/10.3390/s25226988
The paper
Guo, P., & Nakayama, M. (2025). Towards User-Generalizable Wearable-Sensor-Based Human Activity Recognition: A Multi-Task Contrastive Learning Approach. Sensors. https://doi.org/10.3390/s25226988

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.