CoChat

Deep Learning in Human Activity Recognition with Wearable Sensors: A Review on Advances

Open access

The DOI matches its Crossref record; 334 of the 353 references it cites were found in a registry or an index, and 1 are not checked yet; 24 of the 191 sources its claims cite were checked for support.

Abstract

Mobile and wearable devices have enabled numerous applications, including activity tracking, wellness monitoring, and human–computer interaction, that measure and improve our daily lives. Many of these applications are made possible by leveraging the rich collection of low-power sensors found in many mobile and wearable devices to perform human activity recognition (HAR). Recently, deep learning has greatly pushed the boundaries of HAR on mobile and wearable devices. This paper systematically categorizes and summarizes existing work that introduces deep learning methods for wearables-based HAR and provides a comprehensive analysis of the current advancements, developing trends, and major challenges. We also present cutting-edge frontiers and future directions for deep learning-based HAR.

Verified 2026-10-07 · 1 reference not checked yet

Overview

CoChat Summary

TL;DR

This review systematically categorizes deep learning approaches for wearable-sensor human activity recognition (HAR), covering applications, sensors, datasets, algorithms, and open challenges. It concludes that deep learning has advanced HAR but that data, label, modeling, and deployment challenges remain.

Key findings

  • The review screened 8,400 Google Scholar results and ultimately reviewed 176 papers, selecting the top 25% most relevant after exclusions.
  • It organizes wearable HAR deep learning methods into six families—autoencoder, DBN, CNN, RNN (including LSTM/GRU), GAN, and deep reinforcement learning—plus hybrid models, and compares their pros, cons, deployment considerations, and model-selection criteria.
  • Representative reported results include DBN accuracy of 82.3% on Opportunity, 99.2% on USC-HAD, and 99.1% on DSA; a stacked autoencoder reached 92.16% on a UCI HAR dataset; CNN+BiLSTM reached 96.37% on UCI-HAR, 96.05% on WISDM, and 94.29% on PAMAP2.
  • Major challenges and opportunities are grouped into data acquisition (need for more data, quality/missing data, privacy), label acquisition (shortage of labels, in-field datasets), modeling (segmentation, complex activities, generalizability, robustness), and deployment (resource-limited platforms, computation partitioning).

Methods

  • Systematic review following the 27-item PRISMA process; searched Google Scholar with meta-keywords combining HAR/wearable terms and deep learning terms using an AND rule, taking top 200 results per combination.
  • Exclusion criteria removed vision/image/video HAR, environmental-sensor HAR such as WiFi- and RFID-based HAR, minor algorithmic advances, review papers, non-English papers, and works not proposing novel systems or algorithms; final set was 176 papers.
  • Analysis is narrative and taxonomic: categorizes works by sensor, application, deep learning approach, and challenge, and compares algorithms on public datasets, pros/cons, deployment, and high-level model-selection criteria.

Limitations

  • Stated: the review cannot include all published papers in this rapidly growing field; selection of representative works is subject to bias; it may miss the first paper initiating or adopting a method; and heterogeneity in devices, participant demography, and algorithm implementation can affect study results.
  • Evident from the text: as a review, it does not present a new primary experiment or meta-analysis; reliance on Google Scholar relevance and top-25% selection may introduce selection bias; and it excludes image/video and environmental-sensor HAR by design, so findings are scoped to wearable on-body sensing.

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

Integrity signals

Registry

Crossref · 10.3390/s22041476

Cites

334 of 336 references verified0 citation errors · 1 not checkable

Statistics

Nothing to recompute.

Figures

Figure 1 Wearable devices and their application. (a) Distribution of wearable applications [6]. (b) Typical wearable devices. (c) Distribution of wearable devices placed on common body areas [6].
Figure 1 Wearable devices and their application. (a) Distribution of wearable applications [6]. (b) Typical wearable devices. (c) Distribution of wearable devices placed on common body areas [6].CC BY · PubMed Central
Figure 2 Consort diagram outlining how we selected the final papers we included in this work.
Figure 2 Consort diagram outlining how we selected the final papers we included in this work.CC BY · PubMed Central
Figure 3 Taxonomy of Deep Learning-based Human Activity Recognition with Wearables.
Figure 3 Taxonomy of Deep Learning-based Human Activity Recognition with Wearables.CC BY · PubMed Central
Figure 4 Placement of inertial sensors in different datasets: WISDOM; ActRecTut; UCI-HAR; SHO; PAMAP2; and Opportunity.
Figure 4 Placement of inertial sensors in different datasets: WISDOM; ActRecTut; UCI-HAR; SHO; PAMAP2; and Opportunity.CC BY · PubMed Central
Figure 5 Illustration of an autoencoder network [132].
Figure 5 Illustration of an autoencoder network [132].CC BY · PubMed Central
Figure 6 The greedy layer-wise training of DBNs. The first level is trained on triaxial acceleration data. Then, more RBMs are repeatedly stacked to form a deep activity recognition model [151].
Figure 6 The greedy layer-wise training of DBNs. The first level is trained on triaxial acceleration data. Then, more RBMs are repeatedly stacked to form a deep activity recognition model [151].CC BY · PubMed Central
Figure 7 Schematic diagram of an RNN node and LSTM cell [202]. Left: RNN node where ht−1 is the previous hidden state, xt is the current input sample data, ht is the current hidden state, yt is the current output, and ℱ is the activation function. Right: LSTM cell with internal recurrence ct and outer recurrence ht.
Figure 7 Schematic diagram of an RNN node and LSTM cell [202]. Left: RNN node where ht−1 is the previous hidden state, xt is the current input sample data, ht is the current hidden state, yt is the current output, and ℱ is the activation function. Right: LSTM cell with internal recurrence ct and outer recurrence ht.CC BY · PubMed Central
Figure 8 The structure of LSTM and bi-directional LSTM model [204]. (a). LSTM network hidden layers containing LSTM cells and a final softmax layer at the top. (b) bi-directional LSTM network with two parallel tracks in both future (green) and past (red) directions.
Figure 8 The structure of LSTM and bi-directional LSTM model [204]. (a). LSTM network hidden layers containing LSTM cells and a final softmax layer at the top. (b) bi-directional LSTM network with two parallel tracks in both future (green) and past (red) directions.CC BY · PubMed Central
Figure 9 A typical structure of a reinforcement learning network [229].
Figure 9 A typical structure of a reinforcement learning network [229].CC BY · PubMed Central
Figure 10 The structure of generative adversarial network.
Figure 10 The structure of generative adversarial network.CC BY · PubMed Central

Verification

Claims the paper makes

20 Valid · 19 Not valid · 101 Not checked · 99 Own results and interpretations

82 more claims · 8 with their cited sources checked

Create a free account to see all 239 claims and to verify the 167 references still open.

See all claims — free account
  • Probably not a substantial claim. 122
  • It may state two separate claims as one. 58

Create a free account to see these claims.

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

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

Research works printed but not linked: 7

  1. Chen, 2021, NeckFace: Continuously Tracking Full Facial Expressions on Neck-Mounted Wearables, Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., 5, 1
  2. Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., and Bengio, Y. (2014). Generative adversarial nets. Adv. Neural Inf. Process. Syst., 2672–2680. Available online: https://proceedings.neurips.cc/paper/2014/file/5ca3e9b122f61f8f06494c97b1afccf3-Paper.pdf.
  3. Abedin, A., Rezatofighi, H., and Ranasinghe, D.C. (2021). Guided-GAN: Adversarial Representation Learning for Activity Recognition with Wearables. arXiv.
  4. Gulrajani, I., Ahmed, F., Arjovsky, M., Dumoulin, V., and Courville, A. (2017). Improved training of wasserstein gans. arXiv.
  5. Che, T., Li, Y., Jacob, A.P., Bengio, Y., and Li, W. (2016). Mode regularized generative adversarial networks. arXiv.
  6. Rizve, M.N., Duarte, K., Rawat, Y.S., and Shah, M. (2021). In defense of pseudo-labeling: An uncertainty-aware pseudo-label selection framework for semi-supervised learning. arXiv.
  7. Konečnỳ, J., McMahan, B., and Ramage, D. (2015). Federated optimization: Distributed optimization beyond the datacenter. arXiv.

Cited by

See all 625

Metadata

Authors
Zhang, S., Li, Y., Zhang, S., Shahabi, F., Xia, S., Deng, Y., & Alshurafa, N.
Affiliations
  • Northwestern University (US)
  • McGill University (CA)
  • Columbia University (US)
Corresponding author
Zhang, Shibo
Venue
Sensors
Publisher
MDPI AG
Published
2022-02-14
Volume
22
Issue
4
Pages
1476
ISSN
1424-8220 (ISSN-L)
OpenAlex
W3210766530
Cited by
625
References
353
Citations by year
2026: 166 · 2025: 150 · 2024: 135 · 2023: 107 · 2022: 38
FWCI
9.15
Type
journal article
Language
en
Open access
Locations
  • Sensors (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • arXiv (Cornell University) (submittedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
  • arXiv (Cornell University)
Topics
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Context-Aware Activity Recognition Systems
  • Physical Sciences › Engineering › Biomedical Engineering › Gait Recognition and Analysis
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Human Pose and Action Recognition
Keywords
human activity recognition, wearable sensors, deep learning, mobile devices, activity monitoring, wellness monitoring, human-computer interaction, low-power sensors
MeSH terms
Wearable Electronic Devices, Deep Learning, 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 2026-10-07, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-07 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

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

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

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.