CoChat

The DOI matches its Crossref record; 32 of the 32 references it cites were found in a registry or an index; 11 of the 24 sources its claims cite were checked for support.

Adaptive temporal attention mechanism and hybrid deep CNN model for wearable sensor-based human activity recognition

Open access
Wang, Zhixue, Kang, Kai · 2025 · Scientific Reports

Abstract

The recognition of human activity by wearable sensors has garnered significant interest owing to its extensive applications in health, sports, and surveillance systems. This paper presents a novel hybrid deep learning model, termed CNNd-TAm, for the recognition of both basic and complicated activities. The suggested approach enhances spatial feature extraction and long-term temporal dependency modeling by integrating Dilated convolutional networks with a modified temporal attention mechanism. Data from accelerometer and gyroscope sensors in the UTwente dataset, encompassing 13 activities and 10 people, underwent preparation that included filtering, normalization, and the selection of diverse time periods according to the activity type. Experimental findings demonstrate an accuracy of 99.4% in identifying intricate behaviors, such as conversing and consuming coffee, surpassing earlier hybrid deep learning models. This model represents a significant advancement in the development of efficient Human Activity Recognition systems by solving deficiencies in the recognition of intricate activities.

Overview

CoChat Summary

TL;DR

The paper proposes CNNd-TAm, a hybrid deep model combining dilated 1D convolutions, BiGRU, and a modified parallel temporal attention mechanism for wearable-sensor human activity recognition. It reports high accuracy on UTwente, UCI-HAR, and OPPORTUNITY, including 99.4% for complex activities in the abstract and 98.30–99.42% for complex activities in the results.

Key findings

  • On UTwente, tenfold cross-validation for combined activities gave accuracy 99.15 ± 0.25% (CI: [98.9, 99.4]) and F1 98.82 ± 0.28%; for complex activities, CNNd-TAm reached 98.30–99.42%, versus VGG-16 at 94.70–95.50% and ResNet-50 at 95.30–96.20%.
  • UCI-HAR simple/combined activities reached 99.50 ± 0.20% accuracy (CI: [99.3, 99.7]) and F1 99.45 ± 0.22%; OPPORTUNITY was reported at 98.40 ± 0.27% accuracy and F1 98.50 ± 0.25% in the statistical evaluation, with 98.30 ± 0.25% also reported in the confusion-matrix discussion.
  • Training curves showed UTwente final accuracy 99.13% and loss 0.0069, and OPPORTUNITY final accuracy 98.30% and loss 0.022, outperforming CNN-BiGRU, CNN-BiLSTM, CNN-LSTM, and CNN-GRU.
  • The model reduced errors to less than 0.8% when distinguishing similar activities such as drinking and smoking, and showed negligible errors for complex UTwente activities such as conversing and consuming coffee.

Methods

  • Design: CNNd-TAm uses three 1D dilated convolutional layers with dilation rates 1, 2, and 4, max pooling, dropout 0.3, and GeLU activation; a 128-unit BiGRU; and a modified temporal attention mechanism combining parallel Self-Attention and General Attention with an additive connection layer, followed by dense/softmax output and cross-entropy loss. Training used Adam with learning rate 0.001, batch size 64, and 200 epochs.
  • Data: UTwente (13 activities, 10 people, two smartphones at trouser pocket and wrist, accelerometer/gyroscope/linear acceleration/magnetometer, 50 Hz), UCI-HAR (6 activities, 30 people, waist-mounted smartphone, 50 Hz), and OPPORTUNITY (17 activities, 4 people, wearable sensors, 30 Hz). Preprocessing used a third-order Butterworth low-pass filter at 20 Hz, linear interpolation for missing values, Min–Max normalization to [0, 1], activity-specific windows of 20 s for simple, 40 s for complex, and 30 s for mixed activities with 50% overlap, and tenfold cross-validation.
  • Analysis: performance was assessed with accuracy, precision, recall, F1 score, and confusion matrices, comparing CNNd-TAm against VGG-16, ResNet-50, CNN-LSTM, CNN-GRU, CNN-BiLSTM, and CNN-BiGRU.

Limitations

  • Stated: the UTwente dataset has limited user variety and a small sample size (10 users), which may not represent real-world heterogeneity in age, gender, or movement styles; its complex, non-repetitive activities may inadequately represent dynamic real-world conditions.
  • Stated: the authors note reliance on computational complexity and data quality as drawbacks, and recommend future work with multi-objective data such as heart rate or EEG and larger, more diverse datasets.
  • Evident: WISDM performance is only anticipated (>98.5%, SD <0.30%) rather than experimentally validated, and the text reports inconsistent filter/dilation configurations between the architecture description and the hyperparameter section.

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

Integrity signals

Registry

Crossref · 10.1038/s41598-025-18444-6

Cites

32 of 32 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Fig. 1 The proposed framework of S-HAR for CHA recognition.
Fig. 1 The proposed framework of S-HAR for CHA recognition.CC BY-NC-ND · PubMed Central
Fig. 2 Architecture of the proposed CNNd-TAm model.
Fig. 2 Architecture of the proposed CNNd-TAm model.CC BY-NC-ND · PubMed Central
Fig. 3 (a). Temporal Attention Architecture (b). Modified Temporal Attention Architecture.
Fig. 3 (a). Temporal Attention Architecture (b). Modified Temporal Attention Architecture.CC BY-NC-ND · PubMed Central
Fig. 4 (a). GRU recurrent network architecture and (b). Bidirectional GRU.
Fig. 4 (a). GRU recurrent network architecture and (b). Bidirectional GRU.CC BY-NC-ND · PubMed Central
Fig. 5 (a) Self-Attention Network Architecture (b). General-Attention Network Architecture.
Fig. 5 (a) Self-Attention Network Architecture (b). General-Attention Network Architecture.CC BY-NC-ND · PubMed Central
Fig. 6 Confusion matrices for complex activities on UTwente dataset.
Fig. 6 Confusion matrices for complex activities on UTwente dataset.CC BY-NC-ND · PubMed Central
Fig. 7 Confusion matrices for complex activities on OPPORTUNITY dataset.
Fig. 7 Confusion matrices for complex activities on OPPORTUNITY dataset.CC BY-NC-ND · PubMed Central
Fig. 8 Comparison of (a) training accuracy and (b) training loss for Hybrid Deep learning models on the OPPORTUNITY dataset for complex activities over 200 epochs.
Fig. 8 Comparison of (a) training accuracy and (b) training loss for Hybrid Deep learning models on the OPPORTUNITY dataset for complex activities over 200 epochs.CC BY-NC-ND · PubMed Central
Fig. 9 Comparison of (a) training accuracy and (b) training loss for Hybrid Deep learning models on the UTwente dataset for complex activities over 200 epochs.
Fig. 9 Comparison of (a) training accuracy and (b) training loss for Hybrid Deep learning models on the UTwente dataset for complex activities over 200 epochs.CC BY-NC-ND · PubMed Central
Fig. 10 Performance metrics for combined activities across datasets.
Fig. 10 Performance metrics for combined activities across datasets.CC BY-NC-ND · PubMed Central

Verification

Claims the paper makes

6 Valid · 0 Not valid · 8 Not checked · 57 Own results and interpretations

24 more claims

Create a free account to see all 71 claims and to verify the 13 references still open.

See all claims — free account
  • It may state two separate claims as one. 32
  • The quote only partly states this claim. 1
  • Probably not a substantial claim. 10

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.

5 of 5

Citations

Cites

1 retracted

Show all 31 references

Research works printed but not linked: 4

  1. Chen, K. et al. Deep learning for sensor-based human activity recognition: Overview, challenges, and opportunities. ACM Computing Surv. (CSUR) 54(4), 1–40 (2021).
  2. Gupta, S. Deep learning based human activity recognition (HAR) using wearable sensor data. Int. J. Inf. Manag. Data Insights 1(2), 100046 (2021).
  3. Wang, H. et al. Wearable sensor-based human activity recognition using hybrid deep learning techniques. Secur. Commun. Netw. 2020(1), 2132138 (2020).
  4. Athota, R. K. & Sumathi, D. Human activity recognition based on hybrid learning algorithm for wearable sensor data. Int. J. Inf. Technol. 14(7), 3539–3548 (2022).

Cited by

See all 5

Metadata

Authors
Wang, Z., & Kang, K.
Affiliations
Ningxia Normal University (CN)
Corresponding author
Kang, Kai
Venue
Scientific Reports
Published
2025-09-29
Volume
15
Issue
1
Pages
33389-33389
ISSN
2045-2322 (ISSN-L)
OpenAlex
W4414607031
Cited by
5
References
31
Citations by year
2026: 5
FWCI
1.76
Type
journal article
Language
en
Open access
Licence
Locations
  • Scientific Reports (publishedVersion, cc-by-nc-nd)
  • Europe PMC (PubMed Central) (publishedVersion, cc-by-nc-nd)
  • PubMed (publishedVersion)
  • PubMed Central (publishedVersion, cc-by-nc-nd)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
Natural Science Foundation of Ningxia Province (2023AAC03333)
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, temporal attention, dilated convolutional neural network, hybrid CNN architecture, accelerometer and gyroscope data, long-range temporal dependencies, spatial feature extraction
MeSH terms
Wearable Electronic Devices, Deep Learning, Algorithms, Humans, Human Activities, Neural Networks, Computer, Accelerometry

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 2026-10-11, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-11 (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: Adaptive temporal attention mechanism and hybrid deep CNN model for wearable sensor-based human activity recognition. https://app.cochat.ai/papers/10.1038/s41598-025-18444-6
The paper
Wang, Z., & Kang, K. (2025). Adaptive temporal attention mechanism and hybrid deep CNN model for wearable sensor-based human activity recognition. Scientific Reports. https://doi.org/10.1038/s41598-025-18444-6

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.