pith. sign in

arxiv: 2606.13194 · v1 · pith:LK6CN2OJnew · submitted 2026-06-11 · 💻 cs.LG

WHAR Arena: Benchmarking the State of the Art in Efficient Wearable Human Activity Recognition

Maximilian Burzer , Tobias King , Till Riedel , Michael Beigl , Tobias R\"oddiger This is my paper

Pith reviewed 2026-06-27 07:07 UTC · model grok-4.3

classification 💻 cs.LG
keywords wearable human activity recognitionbenchmarkdeep learningdeployment efficiencycross-subject evaluationPareto frontierdomain adaptationon-device metrics
0
0 comments X

The pith

Standardized benchmark of 17 models on 30 datasets shows wearable activity recognition performance has plateaued while efficiency trade-offs stay open.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The authors address inconsistent reporting in wearable human activity recognition by building a unified benchmark that processes 30 datasets the same way and tests models under one cross-subject protocol. They evaluate 17 architectures in thousands of runs, tracking both accuracy and real device costs such as latency and memory. The results indicate that leading models perform similarly with no single winner, but smaller neural networks and random forests achieve the best balance of accuracy and hardware demands. Larger recurrent or hybrid models add cost without extra accuracy. This points to limited further gains from raw predictive power and more room in making models efficient to run and robust across different users or settings.

Core claim

The state of the art in wearable human activity recognition is distributed rather than dominated by a single architecture. CNN-HAR records the highest mean macro-F1, yet top models cluster tightly near a performance ceiling. When deployment costs are considered, compact neural models such as TinierHAR and classical random forests mark the practical Pareto frontier, while larger recurrent and hybrid models incur high hardware costs without matching gains. Consequently, predictive performance has plateaued, but substantial potential remains in optimizing deployment efficiency and improving adaptation to domain shifts.

What carries the argument

The WHAR Arena benchmark that combines 30 datasets under standardized processing, unified model interfaces, and a shared cross-subject evaluation protocol while jointly measuring predictive performance with on-device latency, peak memory, and model size.

If this is right

  • No single architecture dominates predictive performance across the tested datasets.
  • Contemporary models have converged near a predictive performance ceiling.
  • Compact models such as TinierHAR and random forests define the relevant efficiency-accuracy trade-off.
  • Larger recurrent and hybrid models add hardware costs without corresponding accuracy improvements.
  • Future gains are more likely from deployment efficiency improvements and domain-shift adaptation than from further accuracy increases.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Standardized multi-metric benchmarks could reduce wasted effort on incomparable claims in other sensor-based recognition tasks.
  • Hardware measurements taken on a single Android device may understate variation across different wearable processors.
  • Releasing the full evaluation framework allows direct addition of new datasets or metrics by other researchers.
  • Domain-shift robustness could be tested more explicitly by holding out entire user groups or sensor placements not covered in the current protocol.

Load-bearing premise

The 30 datasets together with the single cross-subject protocol and standardized processing give a representative and unbiased picture of real-world performance.

What would settle it

A new architecture that exceeds the current tight performance cluster by a clear margin in macro-F1 while keeping latency and memory below the compact-model frontier on the same Android reference device would disprove the plateau claim.

Figures

Figures reproduced from arXiv: 2606.13194 by Maximilian Burzer, Michael Beigl, Till Riedel, Tobias King, Tobias R\"oddiger.

Figure 1
Figure 1. Figure 1: Public Wearable Human Activity Recognition (WHAR) datasets are highly heterogeneous in terms of file formats, structure, and [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: ACM Digital Library search query used to identify WHAR model papers. [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: A categorical overview of the 30 selected datasets. The bar charts highlight the heterogeneity of our dataset suite along different [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: A multi-dimensional overview of the 30 selected datasets. The chart highlights the diversity of the suite by mapping the number of [PITH_FULL_IMAGE:figures/full_fig_p011_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Entity-relationship diagram illustrating the metadata schema for the standardized data format. [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Model–dataset predictive performance matrix with datasets as rows and models as columns. Each cell reports the mean test macro-F1 [PITH_FULL_IMAGE:figures/full_fig_p014_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Aggregate line-and-range summary plots derived from [PITH_FULL_IMAGE:figures/full_fig_p015_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Pareto-front deployment trade-off summaries for benchmark models with available runtime measurements and exported artifacts. [PITH_FULL_IMAGE:figures/full_fig_p017_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Efficiency-index rankings for the same deployment dimensions shown in Figure [PITH_FULL_IMAGE:figures/full_fig_p018_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Per-dataset latency matrix. Rows correspond to datasets, columns correspond to models, and the lowest latency per dataset is [PITH_FULL_IMAGE:figures/full_fig_p025_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Per-dataset memory matrix. Rows correspond to datasets, columns correspond to models, and the lowest memory overhead per [PITH_FULL_IMAGE:figures/full_fig_p026_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Per-dataset FLOPs matrix with datasets as rows and models as columns. The lowest forward-pass compute per dataset is highlighted. [PITH_FULL_IMAGE:figures/full_fig_p027_12.png] view at source ↗
read the original abstract

Deep learning has become the dominant paradigm in Wearable Human Activity Recognition (WHAR), yet progress is obscured by a comparability crisis. Results are often reported using inconsistent datasets, custom data processing, and varying evaluation protocols, making state-of-the-art claims fragile. We address this with a large-scale, open-source benchmark that integrates 30 diverse datasets under standardized processing, unified model interfaces, and a shared cross-subject evaluation protocol. Evaluating 17 representative architectures across 4760 training runs, we jointly measure predictive performance alongside on-device latency, peak memory, and model size on an Android reference device. Our results reveal that the WHAR state of the art is distributed rather than dominated by a single architecture. While CNN-HAR achieves the highest mean macro-F1, top-performing models cluster tightly, indicating contemporary architectures have converged near a predictive performance ceiling. When accounting for deployment efficiency, compact neural models, such as TinierHAR, and classical Random Forests define the practically relevant Pareto frontier, whereas larger recurrent and hybrid models incur high hardware costs without corresponding performance gains. Consequently, while predictive performance has plateaued, substantial potential for future progress remains in optimizing deployment efficiency and improving adaptation to domain shifts. We release our full framework to support transparent reuse and extension.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper introduces WHAR Arena, a large-scale benchmark integrating 30 diverse datasets under standardized processing and a unified cross-subject evaluation protocol. It evaluates 17 representative architectures across 4760 training runs, jointly measuring predictive performance (macro-F1) alongside on-device latency, peak memory, and model size on an Android reference device. Results show that top models (e.g., CNN-HAR) cluster tightly with no single dominant architecture, compact models like TinierHAR and Random Forests define the Pareto frontier for efficiency, and larger recurrent/hybrid models incur high hardware costs without proportional gains. The central claim is that predictive performance has plateaued while substantial progress remains possible in deployment efficiency and domain-shift adaptation; the full framework is released open-source.

Significance. If the benchmark protocol and results hold, this addresses the comparability crisis in WHAR by establishing a reusable, standardized evaluation framework at a scale (4760 runs) that enables direct comparison of accuracy and hardware metrics. The open-source release and focus on practical Pareto trade-offs are strengths that could guide future work away from accuracy-only optimization. The empirical nature with direct measurements (no circular derivations) adds to its utility as a reference.

major comments (2)
  1. [Abstract and §4 (Results)] Abstract and §4 (Results): the claim that top-performing models 'cluster tightly' and have 'converged near a predictive performance ceiling' is load-bearing for the plateau conclusion, yet no variance measures, confidence intervals, or statistical tests (e.g., paired t-tests or ANOVA across the 4760 runs) are reported to confirm that differences between CNN-HAR and other top models are insignificant.
  2. [§3 (Methods/Datasets)] §3 (Methods/Datasets): the unified cross-subject protocol and selection of the 30 datasets are central to the generalizability of both the plateau and Pareto-frontier claims, but details on data exclusion rules, inclusion criteria, and verification that the protocol avoids bias across domains are insufficient to fully assess representativeness.
minor comments (2)
  1. [Abstract] Abstract: specify how macro-F1 is aggregated (per-dataset then averaged, or pooled) and whether class imbalance handling is uniform across all 30 datasets.
  2. [Figures/Tables] Figure/Table captions (throughout): ensure all hardware metrics explicitly reference the Android device model and measurement methodology for reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive comments, which help improve the statistical support for our claims and the transparency of our dataset protocol. We address each major comment below and will incorporate revisions to strengthen the manuscript.

read point-by-point responses

  1. Referee: [Abstract and §4 (Results)] Abstract and §4 (Results): the claim that top-performing models 'cluster tightly' and have 'converged near a predictive performance ceiling' is load-bearing for the plateau conclusion, yet no variance measures, confidence intervals, or statistical tests (e.g., paired t-tests or ANOVA across the 4760 runs) are reported to confirm that differences between CNN-HAR and other top models are insignificant.

    Authors: We agree that variance measures and statistical tests would strengthen the evidence for the tight clustering of top models and the plateau conclusion. In the revised manuscript, we will report standard deviations across the multiple training runs for macro-F1 scores of the top models, include confidence intervals where appropriate, and add results from paired statistical tests (e.g., paired t-tests or Wilcoxon signed-rank tests) comparing CNN-HAR against other leading models to assess whether differences are statistically insignificant. These additions will be placed in §4 and referenced in the abstract. revision: yes

  2. Referee: [§3 (Methods/Datasets)] §3 (Methods/Datasets): the unified cross-subject protocol and selection of the 30 datasets are central to the generalizability of both the plateau and Pareto-frontier claims, but details on data exclusion rules, inclusion criteria, and verification that the protocol avoids bias across domains are insufficient to fully assess representativeness.

    Authors: We acknowledge that expanded details on dataset curation and protocol safeguards would better allow readers to evaluate representativeness. In the revised §3, we will add explicit inclusion criteria (e.g., minimum subject count, sensor types, activity granularity), data exclusion rules applied during standardization (e.g., removal of incomplete recordings or incompatible label sets), and additional verification steps such as domain coverage analysis and checks for protocol-induced bias across sensor modalities and populations. A new table or subsection will summarize these criteria. revision: yes

Circularity Check

0 steps flagged

No significant circularity identified

full rationale

This paper is a purely empirical benchmark reporting direct measurements of accuracy, latency, memory, and model size across 17 architectures and 30 datasets under a fixed protocol. No derivations, equations, fitted parameters presented as predictions, or self-citation chains appear in the abstract or described methodology. The central claims (performance plateau, efficiency Pareto frontier) follow from observed experimental outcomes rather than reducing to inputs by construction. The analysis is self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claims rest on the domain assumption that the benchmark protocol fairly represents the field; no free parameters or invented entities are introduced.

axioms (1)
  • domain assumption The selected 30 datasets and unified cross-subject protocol fairly represent real-world WHAR variability
    This assumption underpins the generalizability of the plateau and Pareto frontier claims.

pith-pipeline@v0.9.1-grok · 5767 in / 1170 out tokens · 41012 ms · 2026-06-27T07:07:20.192717+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

118 extracted references · 93 canonical work pages · 2 internal anchors

  1. [1]

    Martín Abadi, Ashish Agarwal, Paul Barham, Eugene Brevdo, Zhifeng Chen, Craig Citro, Greg S. Corrado, Andy Davis, Jeffrey Dean, Matthieu Devin, Sanjay Ghemawat, Ian Goodfellow, Andrew Harp, Geoffrey Irving, Michael Isard, Yangqing Jia, Rafal Jozefowicz, Lukasz Kaiser, Manjunath Kudlur, Josh Levenberg, Dan Mané, Rajat Monga, Sherry Moore, Derek Murray, Chr...

    2015

  2. [2]

    Reem Abdel-Salam, Rana Mostafa, and Mayada Hadhood. 2021. Human Activity Recognition using Wearable Sensors: Review, Challenges, Evaluation Benchmark. Vol. 1370. 1–15. doi:10.1007/978-981-16-0575-8_1 arXiv:2101.01665 [cs]

    work page doi:10.1007/978-981-16-0575-8_1 2021

  3. [3]

    Ranasinghe

    Alireza Abedin, Mahsa Ehsanpour, Qinfeng Shi, Hamid Rezatofighi, and Damith C. Ranasinghe. 2021. Attend and Discriminate: Beyond the State-of-the-Art for Human Activity Recognition Using Wearable Sensors. (2021). doi:10.1145/3448083 WHAR Arena: Benchmarking the State of the Art in Efficient Wearable Human Activity Recognition 21

    work page doi:10.1145/3448083 2021

  4. [4]

    Gulzar Alam, Ian McChesney, Peter Nicholl, and Joseph Rafferty. 2023. Open Datasets in Human Activity Recognition Research—Issues and Challenges: A Review.IEEE Sensors Journal23, 22 (Nov. 2023), 26952–26980. doi:10.1109/jsen.2023.3317645 Publisher: Institute of Electrical and Electronics Engineers (IEEE)

    work page doi:10.1109/jsen.2023.3317645 2023

  5. [5]

    Kerem Altun, Billur Barshan, and Orkun Tunçel. 2010. Comparative study on classifying human activities with miniature inertial and magnetic sensors. Pattern Recognition43, 10 (Oct. 2010), 3605–3620. doi:10.1016/j.patcog.2010.04.019

    work page doi:10.1016/j.patcog.2010.04.019 2010

  6. [6]

    Davide Anguita, Alessandro Ghio, Luca Oneto, Xavier Parra, and Jorge L Reyes-Ortiz. 2013. A Public Domain Dataset for Human Activity Recognition Using Smartphones.Computational Intelligence(2013)

    2013

  7. [7]

    Kjærnli, Sverre Herland, Aleksej Logacjov, and Paul Jarle Mork

    Kerstin Bach, Atle Kongsvold, Hilde Bårdstu, Ellen Marie Bardal, Håkon S. Kjærnli, Sverre Herland, Aleksej Logacjov, and Paul Jarle Mork. 2022. A Machine Learning Classifier for Detection of Physical Activity Types and Postures During Free-Living.Journal for the Measurement of Physical Behaviour5, 1 (March 2022), 24–31. doi:10.1123/jmpb.2021-0015

    work page doi:10.1123/jmpb.2021-0015 2022

  8. [8]

    2014.mHealthDroid: A Novel Framework for Agile Development of Mobile Health Applications

    Oresti Banos, Rafael García, Juan Holgado-Terriza, Miguel Damas, Hector Pomares, Ignacio Rojas, and Alejandro Saez. 2014.mHealthDroid: A Novel Framework for Agile Development of Mobile Health Applications. Vol. 8868. doi:10.1007/978-3-319-13105-4_14 Pages: 98

    work page doi:10.1007/978-3-319-13105-4_14 2014

  9. [9]

    Oresti Banos, Mate Toth, Miguel Damas, Hector Pomares, and Ignacio Rojas. 2014. Dealing with the Effects of Sensor Displacement in Wearable Activity Recognition.Sensors14, 6 (June 2014), 9995–10023. doi:10.3390/s140609995

    work page doi:10.3390/s140609995 2014

  10. [10]

    Ganapati Bhat, Nicholas Tran, Holly Shill, and Umit Y. Ogras. 2020. w-HAR: An Activity Recognition Dataset and Framework Using Low-Power Wearable Devices.Sensors20, 18 (Sept. 2020), 5356. doi:10.3390/s20185356

    work page doi:10.3390/s20185356 2020

  11. [11]

    Sizhen Bian, Mengxi Liu, Vitor Fortes Rey, Daniel Geißler, and Paul Lukowicz. 2025. TinierHAR: Towards Ultra-Lightweight Deep Learning Models for Efficient Human Activity Recognition on Edge Devices. InProceedings of the 2025 ACM International Symposium on Wearable Computers (Iswc ’25). Association for Computing Machinery, New York, NY, USA, 163–169. doi:...

    work page doi:10.1145/3715071.3750410 2025

  12. [12]

    Vishwanath Bijalwan, Vijay Bhaskar Semwal, and Vishal Gupta. 2022. Wearable Sensor-Based Pattern Mining for Human Activity Recognition: Deep Learning Approach.Industrial Robot: the international journal of robotics research and application49, 1 (Jan. 2022), 21–33. doi:10.1108/IR-09-2020-0187

    work page doi:10.1108/ir-09-2020-0187 2022

  13. [13]

    Marius Bock, Alexander Hölzemann, Michael Moeller, and Kristof Van Laerhoven. 2021. Improving deep learning for HAR with shallow LSTMs. In Proceedings of the 2021 ACM International Symposium on Wearable Computers. 7–12

    2021

  14. [14]

    Marius Bock, Hilde Kuehne, Kristof Van Laerhoven, and Michael Moeller. 2024. WEAR: An Outdoor Sports Dataset for Wearable and Egocentric Activity Recognition.Proc. ACM Interact. Mob. Wearable Ubiquitous Technol.8, 4 (Nov. 2024), 175:1–175:21. doi:10.1145/3699776

    work page doi:10.1145/3699776 2024

  15. [15]

    Andreas Bulling, Ulf Blanke, and Bernt Schiele. 2014. A tutorial on human activity recognition using body-worn inertial sensors.ACM Computing Surveys (CSUR)46, 3 (2014), 1–33

    2014

  16. [16]

    Maximilian Burzer, Tobias King, Till Riedel, Michael Beigl, and Tobias Röddiger. 2025. WHAR Datasets: An Open Source Library for Wearable Human Activity Recognition. InCompanion of the 2025 ACM International Joint Conference on Pervasive and Ubiquitous Computing. 1315–1322

    2025

  17. [17]

    Santoyo-Ramón, and Jose M

    Eduardo Casilari, Jose A. Santoyo-Ramón, and Jose M. Cano-García. 2017. UMAFall: A Multisensor Dataset for the Research on Automatic Fall Detection. Procedia Computer Science110 (2017), 32–39. doi:10.1016/j.procs.2017.06.110

    work page doi:10.1016/j.procs.2017.06.110 2017

  18. [18]

    Shing Chan, Yuan Hang, Catherine Tong, Aidan Acquah, Abram Schonfeldt, Jonathan Gershuny, and Aiden Doherty. 2024. CAPTURE-24: A large dataset of wrist-worn activity tracker data collected in the wild for human activity recognition.Scientific Data11, 1 (Oct. 2024), 1135. doi:10.1038/s41597-024-03960-3

    work page doi:10.1038/s41597-024-03960-3 2024

  19. [19]

    Chen Chen, Roozbeh Jafari, and Nasser Kehtarnavaz. 2015. UTD-MHAD: A multimodal dataset for human action recognition utilizing a depth camera and a wearable inertial sensor. In2015 IEEE International Conference on Image Processing (ICIP). 168–172. doi:10.1109/ICIP.2015.7350781

    work page doi:10.1109/icip.2015.7350781 2015

  20. [20]

    Kaixuan Chen, Dalin Zhang, Lina Yao, Bin Guo, Zhiwen Yu, and Yunhao Liu. 2021. Deep Learning for Sensor-based Human Activity Recognition: Overview, Challenges, and Opportunities.ACM Comput. Surv.54, 4 (May 2021), 77:1–77:40. doi:10.1145/3447744

    work page doi:10.1145/3447744 2021

  21. [21]

    Roman Chereshnev and Attila Kertesz-Farkas. 2017. HuGaDB: Human Gait Database for Activity Recognition from Wearable Inertial Sensor Networks. doi:10.48550/arXiv.1705.08506 arXiv:1705.08506 [cs]

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1705.08506 2017

  22. [22]

    Stefano Chessa and Stefan Knauth. 2012. Evaluating AAL systems through competitive benchmarking.Proc. Evaluating AAL Systems Through Competitive Benchmarking. Indoor Localization and Tracking309 (2012), 1–13

    2012

  23. [23]

    Maximilian Christ, Nils Braun, Julius Neuffer, and Andreas W Kempa-Liehr. 2018. Time series feature extraction on basis of scalable hypothesis tests (tsfresh–a python package).Neurocomputing307 (2018), 72–77

    2018

  24. [24]

    Mathias Ciliberto, Vitor Fortes Rey, Alberto Calatroni, Paul Lukowicz, and Daniel Roggen. 2021. Opportunity++: A Multimodal Dataset for Video- and Wearable, Object and Ambient Sensors-Based Human Activity Recognition.Frontiers in Computer Science3 (Dec. 2021), 792065. doi:10.3389/fcomp.2021. 792065

    work page doi:10.3389/fcomp.2021 2021

  25. [25]

    Dima Damen. 2014. Scaling Egocentric Vision: The EPIC-KITCHENS Dataset. (2014)

    2014

  26. [26]

    2019.Challenges in Sensor-based Human Activity Recognition and a Comparative Analysis of Benchmark Datasets: A Review

    Anindya Das Antar, Masud Ahmed, and Md Atiqur Rahman Ahad. 2019.Challenges in Sensor-based Human Activity Recognition and a Comparative Analysis of Benchmark Datasets: A Review. 139 pages. doi:10.1109/ICIEV.2019.8858508

    work page doi:10.1109/iciev.2019.8858508 2019

  27. [27]

    Emiro De-La-Hoz-Franco, Paola Ariza-Colpas, Javier Medina Quero, and Macarena Espinilla. 2018. Sensor-Based Datasets for Human Activity Recognition – A Systematic Review of Literature.IEEE Access6 (2018), 59192–59210. doi:10.1109/ACCESS.2018.2873502

    work page doi:10.1109/access.2018.2873502 2018

  28. [28]

    Joseph DelPreto, Chao Liu, Yiyue Luo, Michael Foshey, Yunzhu Li, Antonio Torralba, Wojciech Matusik, and Daniela Rus. [n. d.]. ActionSense: A Multimodal Dataset and Recording Framework for Human Activities Using Wearable Sensors in a Kitchen Environment. ([n. d.])

  29. [29]

    Schmidt, and Geoffrey I

    Angus Dempster, Navid Mohammadi Foumani, Chang Wei Tan, Lynn Miller, Amish Mishra, Mahsa Salehi, Charlotte Pelletier, Daniel F. Schmidt, and Geoffrey I. Webb. 2025. MONSTER: Monash Scalable Time Series Evaluation Repository. doi:10.48550/arXiv.2502.15122 arXiv:2502.15122 [cs]

    work page doi:10.48550/arxiv.2502.15122 2025

  30. [30]

    Florenc Demrozi, Graziano Pravadelli, Azra Bihorac, and Parisa Rashidi. 2020. Human Activity Recognition Using Inertial, Physiological and Environmental Sensors: A Comprehensive Survey.IEEE Access8 (2020), 210816–210836. doi:10.1109/ACCESS.2020.3037715

    work page doi:10.1109/access.2020.3037715 2020

  31. [31]

    Daniel Garcia-Gonzalez, Daniel Rivero, Enrique Fernandez-Blanco, and Miguel R. Luaces. 2020. A Public Domain Dataset for Real-Life Human Activity Recognition Using Smartphone Sensors.Sensors20, 8 (April 2020), 2200. doi:10.3390/s20082200

    work page doi:10.3390/s20082200 2020

  32. [32]

    Daniel Geissler, Dominique Nshimyimana, Vitor Fortes Rey, Sungho Suh, Bo Zhou, and Paul Lukowicz. 2024. Beyond Confusion: A Fine-grained Dialectical Examination of Human Activity Recognition Benchmark Datasets. arXiv:2412.09037 [cs] doi:10.48550/arXiv.2412.09037

    work page doi:10.48550/arxiv.2412.09037 2024

  33. [33]

    Hristijan Gjoreski, Mathias Ciliberto, Lin Wang, Francisco Javier Ordonez Morales, Sami Mekki, Stefan Valentin, and Daniel Roggen. 2018. The University of Sussex-Huawei Locomotion and Transportation Dataset for Multimodal Analytics With Mobile Devices.IEEE Access6 (2018), 42592–42604. doi:10.1109/ACCESS.2018.2858933

    work page doi:10.1109/access.2018.2858933 2018

  34. [34]

    Xingyu Gong, Xinyang Zhang, and Na Li. 2024. Lightweight Human Activity Recognition Method Based on the MobileHARC Model.Systems Science & Control Engineering12, 1 (Dec. 2024), 2328549. doi:10.1080/21642583.2024.2328549

    work page doi:10.1080/21642583.2024.2328549 2024

  35. [35]

    Kristen Grauman, Andrew Westbury, Eugene Byrne, Zachary Chavis, Antonino Furnari, Rohit Girdhar, Jackson Hamburger, Hao Jiang, Miao Liu, Xingyu Liu, et al. 2022. Ego4d: Around the world in 3,000 hours of egocentric video. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition. 18995–19012. 22 Burzer, King et al

    2022

  36. [36]

    Fuqiang Gu, Mu-Huan Chung, Mark Chignell, Shahrokh Valaee, Baoding Zhou, and Xue Liu. 2021. A Survey on Deep Learning for Human Activity Recognition.ACM Comput. Surv.54, 8 (Oct. 2021), 177:1–177:34. doi:10.1145/3472290

    work page doi:10.1145/3472290 2021

  37. [37]

    Yu Guan and Thomas Plötz. 2017. Ensembles of Deep LSTM Learners for Activity Recognition Using Wearables.Proc. ACM Interact. Mob. Wearable Ubiquitous Technol.1, 2 (June 2017), 11:1–11:28. doi:10.1145/3090076

    work page doi:10.1145/3090076 2017

  38. [38]

    Moritz Hardt. 2025. The Emerging Science of Machine Learning Benchmarks. Online at https://mlbenchmarks.org. Manuscript

    2025

  39. [39]

    Alexander Hoelzemann, Julia Lee Romero, Marius Bock, Kristof Van Laerhoven, and Qin Lv. 2023. Hang-Time HAR: A Benchmark Dataset for Basketball Activity Recognition Using Wrist-Worn Inertial Sensors.Sensors23, 13 (June 2023), 5879. doi:10.3390/s23135879

    work page doi:10.3390/s23135879 2023

  40. [40]

    Md Meem Hossain, The Anh Han, Safina Showkat Ara, and Zia Ush Shamszaman. 2025. Benchmarking Classical, Deep, and Generative Models for Human Activity Recognition. arXiv:2501.08471 [cs] doi:10.48550/arXiv.2501.08471

    work page doi:10.48550/arxiv.2501.08471 2025

  41. [41]

    Black, Otmar Hilliges, and Gerard Pons-Moll

    Yinghao Huang, Manuel Kaufmann, Emre Aksan, Michael J. Black, Otmar Hilliges, and Gerard Pons-Moll. 2018. Deep Inertial Poser: Learning to Reconstruct Human Pose from Sparse Inertial Measurements in Real Time.ACM Trans. Graph.37, 6 (Dec. 2018), 185:1–185:15. doi:10.1145/3272127.3275108

    work page doi:10.1145/3272127.3275108 2018

  42. [42]

    Yiran Huang, Haibin Zhao, Yexu Zhou, Till Riedel, and Michael Beigl. 2024. Standardizing Your Training Process for Human Activity Recognition Models: A Comprehensive Review in the Tunable Factors. doi:10.48550/arXiv.2401.05477 arXiv:2401.05477 [cs]

    work page doi:10.48550/arxiv.2401.05477 2024

  43. [43]

    Yiran Huang, Yexu Zhou, Haibin Zhao, Till Riedel, and Michael Beigl. 2024. A Survey on Wearable Human Activity Recognition: Innovative Pipeline Development for Enhanced Research and Practice. In2024 International Joint Conference on Neural Networks (IJCNN). 1–10. doi:10.1109/IJCNN60899.2024. 10650060

    work page doi:10.1109/ijcnn60899.2024 2024

  44. [44]

    Jonas Hummel, Tobias Röddiger, Valeria Zitz, Philipp Lepold, Michael Küttner, Marius Prill, Christopher Clarke, Hans Gellersen, and Michael Beigl. 2025. EarXplore: An Open Research Database on Earable Interaction. arXiv:2507.20656 [cs.HC] https://arxiv.org/abs/2507.20656

    arXiv 2025

  45. [45]

    Saurav Jha, Martin Schiemer, Franco Zambonelli, and Juan Ye. 2021. Continual Learning in Sensor-Based Human Activity Recognition: An Empirical Benchmark Analysis.Information Sciences575 (Oct. 2021), 1–21. doi:10.1016/j.ins.2021.04.062

    work page doi:10.1016/j.ins.2021.04.062 2021

  46. [46]

    Human Activity Recognition Based on Wearable Sensor Data: A Standardization of the State-of-the-Art

    Artur Jordao, Antonio C. Nazare, Jessica Sena, and William Robson Schwartz. 2019. Human Activity Recognition Based on Wearable Sensor Data: A Standardization of the State-of-the-Art. doi:10.48550/arXiv.1806.05226 arXiv:1806.05226 [cs]

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1806.05226 2019

  47. [47]

    Misha Karim, Shah Khalid, Aliya Aleryani, Jawad Khan, Irfan Ullah, and Zafar Ali. 2024. Human Action Recognition Systems: A Review of the Trends and State-of-the-Art.IEEE Access12 (2024), 36372–36390. doi:10.1109/ACCESS.2024.3373199

    work page doi:10.1109/access.2024.3373199 2024

  48. [48]

    Nobuo Kawaguchi, Nobuhiro Ogawa, Yohei Iwasaki, Katsuhiko Kaji, Tsutomu Terada, Kazuya Murao, Sozo Inoue, Yoshihiro Kawahara, Yasuyuki Sumi, and Nobuhiko Nishio. 2011. Hasc challenge: gathering large scale human activity corpus for the real-world activity understandings. InProceedings of the 2nd augmented human international conference. 1–5

    2011

  49. [49]

    Quan Kong, Ziming Wu, Ziwei Deng, Martin Klinkigt, Bin Tong, and Tomokazu Murakami. 2019. Mmact: A large-scale dataset for cross modal human action understanding. InProceedings of the IEEE/CVF International Conference on Computer Vision. 8658–8667

    2019

  50. [50]

    Pranjal Kumar, Siddhartha Chauhan, and Lalit Kumar Awasthi. 2024. Human Activity Recognition (HAR) Using Deep Learning: Review, Methodologies, Progress and Future Research Directions.Archives of Computational Methods in Engineering31, 1 (Jan. 2024), 179–219. doi:10.1007/s11831-023-09986-x

    work page doi:10.1007/s11831-023-09986-x 2024

  51. [51]

    Kwapisz, Gary M

    Jennifer R. Kwapisz, Gary M. Weiss, and Samuel A. Moore. 2011. Activity recognition using cell phone accelerometers.SIGKDD Explor. Newsl.12, 2 (March 2011), 74–82. doi:10.1145/1964897.1964918

    work page doi:10.1145/1964897.1964918 2011

  52. [52]

    Tianzheng Liao, Jinjin Zhao, Yushi Liu, Kamen Ivanov, Jing Xiong, and Yan Yan. 2022. Deep Transfer Learning with Graph Neural Network for Sensor-Based Human Activity Recognition. In2022 IEEE International Conference on Bioinformatics and Biomedicine (BIBM). 2445–2452. doi:10.1109/BIBM55620.2022. 9995660

    work page doi:10.1109/bibm55620.2022 2022

  53. [53]

    Hui Liu, Yale Hartmann, and Tanja Schultz. 2021. CSL-SHARE: A Multimodal Wearable Sensor-Based Human Activity Dataset.Frontiers in Computer Science3 (Oct. 2021), 759136. doi:10.3389/fcomp.2021.759136

    work page doi:10.3389/fcomp.2021.759136 2021

  54. [54]

    Aleksej Logacjov, Kerstin Bach, Atle Kongsvold, Hilde Bremseth Bårdstu, and Paul Jarle Mork. 2021. HARTH: A Human Activity Recognition Dataset for Machine Learning.Sensors21, 23 (Nov. 2021), 7853. doi:10.3390/s21237853

    work page doi:10.3390/s21237853 2021

  55. [55]

    Ilya Loshchilov and Frank Hutter. 2017. Decoupled weight decay regularization.arXiv preprint arXiv:1711.05101(2017)

    Pith/arXiv arXiv 2017

  56. [56]

    Wang Lu, Yao Zhu, and Jindong Wang. 2025. HAROOD: A Benchmark for Out-of-distribution Generalization in Sensor-based Human Activity Recognition. arXiv:2512.10807 [cs] doi:10.48550/arXiv.2512.10807

    work page doi:10.48550/arxiv.2512.10807 2025

  57. [57]

    Tanjid Hasan Tonmoy, Kishor Kumar Bhaumik, A

    Saif Mahmud, M. Tanjid Hasan Tonmoy, Kishor Kumar Bhaumik, A. K. M. Mahbubur Rahman, M. Ashraful Amin, Mohammad Shoyaib, Muhammad Asif Hossain Khan, and Amin Ahsan Ali. 2020. Human Activity Recognition from Wearable Sensor Data Using Self-Attention. arXiv:2003.09018 [cs] doi:10.48550/arXiv.2003.09018

    work page doi:10.48550/arxiv.2003.09018 2020

  58. [58]

    Mohammad Malekzadeh, Richard Clegg, Andrea Cavallaro, and Hamed Haddadi. 2021. Dana: Dimension-adaptive neural architecture for multivariate sensor data.Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies5, 3 (2021), 1–27

    2021

  59. [59]

    Clegg, Andrea Cavallaro, and Hamed Haddadi

    Mohammad Malekzadeh, Richard G. Clegg, Andrea Cavallaro, and Hamed Haddadi. 2019. Mobile sensor data anonymization. InProceedings of the International Conference on Internet of Things Design and Implementation. ACM, Montreal Quebec Canada, 49–58. doi:10.1145/3302505.3310068

    work page doi:10.1145/3302505.3310068 2019

  60. [60]

    Lourdes Martínez-Villaseñor, Hiram Ponce, Jorge Brieva, Ernesto Moya-Albor, José Núñez-Martínez, and Carlos Peñafort-Asturiano. 2019. UP-Fall Detection Dataset: A Multimodal Approach.Sensors19, 9 (April 2019), 1988. doi:10.3390/s19091988

    work page doi:10.3390/s19091988 2019

  61. [61]

    Sakorn Mekruksavanich and Anuchit Jitpattanakul. 2020. Smartwatch-based Human Activity Recognition Using Hybrid LSTM Network. In2020 IEEE SENSORS. 1–4. doi:10.1109/SENSORS47125.2020.9278630

    work page doi:10.1109/sensors47125.2020.9278630 2020

  62. [62]

    Shenghuan Miao, Ling Chen, Rong Hu, and Yingsong Luo. 2022. Towards a Dynamic Inter-Sensor Correlations Learning Framework for Multi-Sensor-Based Wearable Human Activity Recognition. (2022). doi:10.1145/3550331

    work page doi:10.1145/3550331 2022

  63. [63]

    Daniela Micucci, Marco Mobilio, and Paolo Napoletano. 2017. UniMiB SHAR: A Dataset for Human Activity Recognition Using Acceleration Data from Smartphones.Applied Sciences7, 10 (Oct. 2017), 1101. doi:10.3390/app7101101

    work page doi:10.3390/app7101101 2017

  64. [64]

    Moore, Hamish G

    Steven T. Moore, Hamish G. MacDougall, and William G. Ondo. 2008. Ambulatory monitoring of freezing of gait in Parkinson’s disease.Journal of Neuroscience Methods167, 2 (Jan. 2008), 340–348. doi:10.1016/j.jneumeth.2007.08.023

    work page doi:10.1016/j.jneumeth.2007.08.023 2008

  65. [65]

    Seyed Ahmadreza Mousavi and Rastko Selmic. 2023. Wearable Smart Rings for Multifinger Gesture Recognition Using Supervised Learning.IEEE Transactions on Instrumentation and Measurement72 (2023), 1–12. doi:10.1109/TIM.2023.3304703

    work page doi:10.1109/tim.2023.3304703 2023

  66. [66]

    Murahari and Thomas Plötz

    Vishvak S. Murahari and Thomas Plötz. 2018. On Attention Models for Human Activity Recognition. InProceedings of the 2018 ACM International Symposium on Wearable Computers (ISWC ’18). Association for Computing Machinery, New York, NY, USA, 100–103. doi:10.1145/3267242.3267287

    work page doi:10.1145/3267242.3267287 2018

  67. [67]

    Katsuyuki Nakamura, Serena Yeung, Alexandre Alahi, and Li Fei-Fei. 2017. Jointly Learning Energy Expenditures and Activities Using Egocentric Multimodal Signals. In2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, Honolulu, HI, 6817–6826. doi:10.1109/CVPR.2017.721

    work page doi:10.1109/cvpr.2017.721 2017

  68. [68]

    Otávio Napoli, Dami Duarte, Patrick Alves, Darlinne Hubert Palo Soto, Henrique Evangelista De Oliveira, Anderson Rocha, Levy Boccato, and Edson Borin

  69. [69]

    2024), 1192

    A benchmark for domain adaptation and generalization in smartphone-based human activity recognition.Scientific Data11, 1 (Nov. 2024), 1192. doi:10.1038/s41597-024-03951-4 WHAR Arena: Benchmarking the State of the Art in Efficient Wearable Human Activity Recognition 23

    work page doi:10.1038/s41597-024-03951-4 2024

  70. [70]

    Duc-Anh Nguyen and Nhien-An Le-Khac. 2024. SoK: Behind the Accuracy of Complex Human Activity Recognition Using Deep Learning. In2024 International Joint Conference on Neural Networks (IJCNN). 1–8. doi:10.1109/IJCNN60899.2024.10650322

    work page doi:10.1109/ijcnn60899.2024.10650322 2024

  71. [71]

    Friedrich Niemann, Fernando Moya Rueda, Moh’d Khier Al Kfari, Nilah Ravi Nair, Stefan Ludtke, and Alice Kirchheim. [n. d.]. Towards Standardized Dataset Creation for Human Activity Recognition: Framework, Taxonomy, Checklist, and Best Practices. ([n. d.])

  72. [72]

    Pierre-Emmanuel Novac, Alain Pegatoquet, Benoît Miramond, and Christophe Caquineau. 2022. UCA-EHAR: A Dataset for Human Activity Recognition with Embedded AI on Smart Glasses.Applied Sciences12, 8 (April 2022), 3849. doi:10.3390/app12083849

    work page doi:10.3390/app12083849 2022

  73. [73]

    Ferda Ofli, Rizwan Chaudhry, Gregorij Kurillo, René Vidal, and Ruzena Bajcsy. 2013. Berkeley MHAD: A Comprehensive Multimodal Human Action Database. In2013 IEEE Workshop on Applications of Computer Vision (W ACV). 53–60. doi:10.1109/WACV.2013.6474999

    work page doi:10.1109/wacv.2013.6474999 2013

  74. [74]

    Godwin Ogbuabor, Juan Augusto Wrede, Ralph Moseley, and Alechia Van Wyk. 2021. Context-Aware Support for Cardiac Health Monitoring Using Federated Machine Learning. 267–281. doi:10.1007/978-3-030-91100-3_22

    work page doi:10.1007/978-3-030-91100-3_22 2021

  75. [75]

    Kamsiriochukwu Ojiako and Katayoun Farrahi. 2023. MLPs Are All You Need for Human Activity Recognition.Applied Sciences13, 20 (Oct. 2023). doi:10.3390/app132011154

    work page doi:10.3390/app132011154 2023

  76. [76]

    Francisco Javier Ordóñez and Daniel Roggen. 2016. Deep Convolutional and LSTM Recurrent Neural Networks for Multimodal Wearable Activity Recognition.Sensors16, 1 (2016). doi:10.3390/s16010115

    work page doi:10.3390/s16010115 2016

  77. [77]

    Eline Slagboom, Marian Beekman, and Arno Knobbe

    Stylianos Paraschiakos, Cláudio Rebelo De Sá, Jeremiah Okai, P. Eline Slagboom, Marian Beekman, and Arno Knobbe. 2022. A recurrent neural network architecture to model physical activity energy expenditure in older people.Data Mining and Knowledge Discovery36, 1 (Jan. 2022), 477–512. doi:10.1007/s10618-021-00817-w

    work page doi:10.1007/s10618-021-00817-w 2022

  78. [78]

    Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, Alban Desmaison, Andreas Kopf, Edward Yang, Zachary DeVito, Martin Raison, Alykhan Tejani, Sasank Chilamkurthy, Benoit Steiner, Lu Fang, Junjie Bai, and Soumith Chintala. 2019. PyTorch: An Imperative Style, High...

    2019

  79. [79]

    Pedregosa, G

    F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V. Dubourg, J. Vanderplas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot, and E. Duchesnay. 2011. Scikit-Learn: Machine Learning in Python.Journal of Machine Learning Research12 (2011), 2825–2830

    2011

  80. [80]

    PyTorch. [n. d.]. ExecuTorch. https://github.com/pytorch/executorch. Accessed 2026-05-01

    2026

Showing first 80 references.