Ensembles of Deep LSTM Learners for Activity Recognition using Wearables
Open accessThe DOI matches its Crossref record; 53 of the 65 references it cites were found in a registry or an index, and 10 are not checked yet; none of the 62 sources its claims cite is checked for support yet.
Abstract
Recently, deep learning (DL) methods have been introduced very successfully into human activity recognition (HAR) scenarios in ubiquitous and wearable computing. Especially the prospect of overcoming the need for manual feature design combined with superior classification capabilities render deep neural networks very attractive for real-life HAR applications. Even though DL-based approaches now outperform the state-of-the-art in a number of recognition tasks, still substantial challenges remain. Most prominently, issues with real-life datasets, typically including imbalanced datasets and problematic data quality, still limit the effectiveness of activity recognition using wearables. In this paper we tackle such challenges through Ensembles of deep Long Short Term Memory (LSTM) networks. LSTM networks currently represent the state-of-the-art with superior classification performance on relevant HAR benchmark datasets. We have developed modified training procedures for LSTM networks and combine sets of diverse LSTM learners into classifier collectives. We demonstrate that Ensembles of deep LSTM learners outperform individual LSTM networks and thus push the state-of-the-art in human activity recognition using wearables. Through an extensive experimental evaluation on three standard benchmarks (Opportunity, PAMAP2, Skoda) we demonstrate the excellent recognition capabilities of our approach and its potential for real-life applications of human activity recognition.
Checked 2026-10-05 · 10 references not checked yet
Overview
CoChat Summary
TL;DR
The paper introduces ensembles of deep LSTM learners for wearable activity recognition, using an epoch-wise bagging-style training procedure and score-level fusion of multiple LSTM models. Across Opportunity, PAMAP2, and Skoda, the ensembles outperform individual LSTMs and state-of-the-art baselines, with the largest gains on the challenging, imbalanced Opportunity dataset.
Key findings
- Ensembles of LSTM learners improve over single LSTMs on all three benchmarks. For Opportunity, single CE LSTM F1 = 0.674 ± 0.021 vs. Ensemble M=10 CE = 0.716 ± 0.011 and M=20 CE = 0.720 ± 0.011; the best combined CE+F1 ensemble reaches 0.726 ± 0.008. For PAMAP2, single CE = 0.802 ± 0.045 vs. Ensemble M=10 CE = 0.848 ± 0.021 and M=20 CE = 0.852 ± 0.026; combined CE+F1 = 0.854 ± 0.026. For Skoda, single CE = 0.918 ± 0.007 vs. Ensemble M=10 CE = 0.925 ± 0.002 and M=20 CE = 0.926 ± 0.001; combined CE+F1 = 0.924 ± 0.002.
- The LSTM ensemble outperforms reported state-of-the-art baselines: LSTM baseline 0.659/0.756/0.904, DeepConvLSTM 0.672/0.748/0.912, and LSTM-S 0.684/0.838/0.921 for Opportunity/PAMAP2/Skoda, respectively, versus the ensemble’s 0.726/0.854/0.924.
- Performance generally plateaus at M=10 base learners; increasing to M=20 gives no statistically significant improvement except for Opportunity with F1 loss, where the improvement is “just significant.” Combining CE and F1 loss functions further improves Opportunity but not PAMAP2, and Skoda shows no substantial boost.
- The training procedure intentionally omits data: on average about 35%–38% of training data is not used per epoch when using a uniformly distributed random number generator for training frames.
Methods
- The authors use two-layer LSTM networks with 256 LSTM units per layer, dropout 0.5 at both hidden layers, ADAM with learning rate 0.001, implemented in Lasagne/Theano and trained on an NVIDIA Tesla K40. Training uses randomly drawn mini-batch sizes from U(128, 256) and frame lengths from U(16, 32); test-time inference is sample-wise with mini-batch size and frame length set to 1.
- The core method is epoch-wise bagging: for each epoch, random mini-batch size, random starting positions, and random frame lengths are drawn; one LSTM learner is produced per epoch (MaxEpoch = 100). The M best models on a validation set are retained and fused by arithmetic mean of their softmax class-probability vectors. Both cross-entropy and F1 loss are explored.
- Evaluation uses three standard benchmarks: Opportunity, PAMAP2, and Skoda, with hold-out validation/test protocols replicating prior work. Results are mean F1 scores over 30 repetitions, with two-tailed independent t-tests for significance.
Limitations
- The paper states that implementation issues for field deployments or products are not yet addressed; ensembles inflate the parameter space, though the authors argue pruning remains accessible because the base LSTM models are unchanged.
- The paper notes that F1 loss is a global measure and cannot directly measure individual sample losses; because LSTMs are trained on mini-batches, it may not precisely approximate the true gradient toward optimal F1 over the whole training set.
- The paper states that Skoda shows no substantial performance boost, attributed to only one worker being considered and the task lacking challenging test data from different environments; the baseline models already achieve high specialisation (F1 > 0.9). Evident from the text: evaluation is limited to three benchmark datasets and does not include an in-the-wild deployment, and frame-based results from prior work were converted to sample-wise predictions for comparability.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
0 Valid · 11 Not valid · 48 Not checked · 80 Own results and interpretations
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Through an extensive experimental evaluation on three standard benchmarks (Opportunity, PAMAP2, Skoda) we demonstrate the excellent recognition capabilities of our approach and its potential for real-life applications of human activity recognition.”
boosted- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“By outperforming the state-of-the-art our results indicate increased robustness of our Ensembles of deep LSTM learners with regards to typical challenges of human activity recognition in ubiquitous and wearable computing scenarios.”
hedgedstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Rests on an assumption not shown here.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“We demonstrate that Ensembles of deep LSTM learners outperform individual LSTM networks and thus push the state-of-the-art in human activity recognition using wearables.”
boostedstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“strong learners -LSTM models in our case- can indeed be integrated into very effective ensembles that outperform single, that is, individual models. Recognition results are consistent across all three tasks”
boostedstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper states — it cites no source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“PAMAP2 does not benefit from diversity in loss function”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
55 more claims
Create a free account to see all 139 claims and to verify the 62 references still open.
- It may state two separate claims as one. 32
- The quote only partly states this claim. 2
- Probably not a substantial claim. 59
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?
The cited source says what the claim attributes to it.
In the citing paper
“Many wearables datasets exhibit substantial class imbalance [68].”
From Past, Present, and Future of Sensor-based Human Activity Recognition Using Wearables: A Surveying Tutorial on a Still Challenging Task (2025) · cites this paper as [68]
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.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 3.2 Modified Training for LSTM Models · p. 12
“Overall, only a total of ˆ B ⌊ T/ ˆ B ⌋ samples are used in each epoch.”
§ ALGORITHM 1: Epoch-wise bagging for LSTM learners. · p. 13
“while L seq ≤ ⌊ T/ ˆ B j ⌋ do draw random integer ˆ L i (i.e., frame length) from discrete uniform distribution U ( L low , L high ) ;”
Citations
Cites
Breiman, L. (2001) · Machine Learning · doi:10.1023/a:1010933404324 · Cited by 133,538
Hochreiter, S., & Schmidhuber, J. (1997) · Neural Computation · doi:10.1162/neco.1997.9.8.1735 · Cited by 102,318
Adam: A Method for Stochastic Optimization
Kingma, D. P., & Ba, J. (2014) · arXiv · doi:10.48550/arxiv.1412.6980 · Cited by 98,946
LeCun, Y., Bengio, Y., & Hinton, G. (2015) · Nature · doi:10.1038/nature14539 · Cited by 84,126
The Elements of Statistical Learning
Hastie, T., Tibshirani, R., & Friedman, J. (2009) · Springer Series in Statistics · doi:10.1007/978-0-387-84858-7 · Cited by 25,970
The Elements of Statistical Learning
Hastie, T., Friedman, J., & Tibshirani, R. (2001) · Springer Series in Statistics · doi:10.1007/978-0-387-21606-5 · Cited by 24,162
Breiman, L. (1996) · Machine Learning · doi:10.1007/bf00058655 · Cited by 16,325
Parallel Distributed Processing: Explorations in the Microstructure of Cognition: Foundations
Rumelhart, D. E., McClelland, J. L., & AU (1986) · doi:10.7551/mitpress/5236.001.0001 · Cited by 15,686
The Computer for the 21st Century
Weiser, M. (1991) · Scientific American · doi:10.1038/scientificamerican0991-94 · Cited by 10,017
Large-Scale Machine Learning with Stochastic Gradient Descent
Bottou, L. (2010) · Proceedings of COMPSTAT'2010 · doi:10.1007/978-3-7908-2604-3_16 · Cited by 5,795
Research works printed but not linked: 2
- James L McClelland and David E Rumelhart . 1986. Parallel distributed processing . MIT Press . James L McClelland and David E Rumelhart. 1986. Parallel distributed processing. MIT Press.
- Theano Development Team . 2016 . Theano: A Python framework for fast computation of mathematical expressions. arXiv e-prints abs/1605.02688 (May 2016). http://arxiv.org/abs/1605.02688 Theano Development Team. 2016. Theano: A Python framework for fast computation of mathematical expressions. arXiv e-prints abs/1605.02688 (May 2016). http://arxiv.org/abs/1605.02688 · arXiv
Cited by
Items 401–500 of 539
Thomanek, R., Roschke, C., Zimmer, F., Rolletschke, T., Manthey, R., Vodel, M., Platte, B., Heinzig, M., Eibl, M., Hösel, C., et al. (2020) · IEEE/CVF Winter Conference on Applications of Computer Vision Workshops (WACVW) · doi:10.1109/wacvw50321.2020.9096936 · Cited by 3
“While current approaches use tubes and streams in the form of 3D CNNs [4, 16, 8], we focus on [17, 10, 9, 5], which detects activities based on the motion constellation of skeletal points or sensor values using an LSTM architecture.”
MethodDeepER: A Deep Learning based Emergency Resolution Time Prediction System
Bejarano, G., Kulkarni, A., Luo, X., Seetharam, A., & Ramesh, A. (2020) · IEEE International Conference on Internet of Things (iThings) and IEEE Green Computing and Communications (GreenCom) and IEEE Cyber, Physical and Social Computing (CPSCom) and IEEE Smart Data (SmartData) · doi:10.1109/ithings-greencom-cpscom-smartdata-cybermatics50389.2020.00090 · Cited by 3
“With the growth and development of smart cities and cyber-physical systems, a variety of machine learning approaches have been adopted to address different problems in these domains [2], [3], [4], [5], [6], [7].”
BackgroundWan, B., & Shen, Y. C. (2021) · International Conference on Computers in Management and Business (ICCMB) · doi:10.1145/3450588.3450606 · Cited by 3
Automated General Movement Assessment for Perinatal Stroke Screening in Infants
Gao, Y., Long, Y., Guan, Y., Basu, A. P., Baggaley, J., & Plötz, T. (2019) · Computer communications and networks · doi:10.1007/978-3-030-25590-9_8 · Cited by 4
Self-citationPerMTL: A Multi-Task Learning Framework for Skilled Human Performance Assessment
Ghosh, I., Chakma, A., Ramamurthy, S. R., Roy, N., & Waytowich, N. R. (2022) · International Conference on Machine Learning and Applications · doi:10.1109/icmla55696.2022.00177 · Cited by 3
“The overlap-defined windows technique is better for extracting temporal patterns for micro-activities than the activity and event-defined windows [18].”
MethodOn Training Strategies for LSTMs in Sensor-Based Human Activity Recognition
Shao, S., Guan, Y., Guan, X., Missier, P., & Plötz, T. (2023) · IEEE International Conference on Pervasive Computing and Communications Workshops (PerCom Workshops) · doi:10.1109/percomworkshops56833.2023.10150305 · Cited by 3
“2) Network Architectures and Hyper-parameters: For comparison to the unmodified LSTM baseline, we followed [18] and used a two-layer LSTM, with 256 hidden units for each layer.”
BackgroundMethodInfluentialSelf-citationGupta, G., Ramesh, K., Bhattacharya, A., Gupta, D., Sharma, R., Chandran, N., & Sen, R. (2023) · Proceedings on Privacy Enhancing Technologies · doi:10.56553/popets-2023-0118 · Cited by 3
“For example, GCN has been used for text classification [66], image classification [11], recommendation systems [41, 61]; LSTM for keyword spotting [8, 10] and gesture detection and human activity recognition [21, 69]; and GCN-LSTM for travel time estimation [32, 70].”
MethodSakshi, Bhatia, M. P. S., & Chakraborty, P. (2023) · International Journal of Information Technology · doi:10.1007/s41870-023-01484-4 · Cited by 3
“…feature set has led to a significant enhancement in accuracy and f1 score, reaching up to 96.57% and 0.966 respectively when utilizing the RF [2,8,12,13,17,18,19,20,22,24,27,32,34,35,36,37,38,39,40,43,45,46,48,49,50,51,53,54,57,59,60,61,65,66,67,70,76,77,78,82,83,85,86,91,92] { CS ( f )}…”
InfluentialTemporal Differential Privacy for Human Activity Recognition
Roy, D., & Girdzijauskas, Š. (2023) · IEEE International Conference on Data Science and Advanced Analytics (DSAA) · doi:10.1109/dsaa60987.2023.10302475 · Cited by 6
“The extraction of temporal sequences is popularly used in many previous HAR works [11], [16], [27].”
MethodHuman Activity Recognition Using Deep Models and Its Analysis from Domain Adaptation Perspective
Gurov, N., Khan, A., Hussain, R., & Khattak, A. M. (2019) · Lecture notes in computer science · doi:10.1007/978-3-030-29852-4_15 · Cited by 4
A Hybrid Deep Learning Model for Human Activity Recognition Using Wearable Sensors
Gaurav, K., Roy, B., & Bharti, J. (2022) · Algorithms for intelligent systems · doi:10.1007/978-981-16-9650-3_16 · Cited by 4
Shao, S., Guan, Y., Guan, X., Missier, P., & Plötz, T. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2212.13918 · Cited by 4
“For OPP dataset, in the LOSO-CV setting data from all four participants were used; while in the hold-out setting, we used the protocol defined in [21], [23].”
BackgroundMethodInfluentialSelf-citationFullie and Wiselie: A Dual-Stream Recurrent Convolutional Attention Model for Activity RecognitionPreprint
Chen, K., Yao, L., Gu, T., Yu, Z., Wang, X., & Zhang, D. (2017) · arXiv (Cornell University) · doi:10.13140/rg.2.2.21831.39847 · Cited by 3
Arakawa, R., Yakura, H., & Goel, M. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2407.16785 · Cited by 3
“Among many modalities used for HAR, smartwatch-based systems often use audio and motion data [11, 18, 26, 27, 34].”
Teng, T. C. I., Wan, J., & Zhang, X. (2024) · Knowledge and Information Systems · doi:10.1007/s10115-024-02154-y · Cited by 3
“Recurrent neural networks (RNN) and their variants [24, 25] are also widely used for processing time-series data to capture dynamic information over time.”
Human Movement Recognition with Machine Learning Techniques
Sinha, R. K., Sinha, P. C., & Tiwari, M. (2024) · IEEE International Conference on Advances in Computing Research on Science Engineering and Technology (ACROSET) · doi:10.1109/acroset62108.2024.10743652 · Cited by 2
Hybrid Neuromorphic‐Federated Learning for Activity Recognition Using Multi‐modal Wearable Sensors
Khan, A. R., Manzoor, H. U., Ayaz, F., Imran, M. A., & Zoha, A. (2024) · Wiley eBooks · doi:10.1002/9781394257744.ch6 · Cited by 2
Unsupervised Deep Clustering for Human Behavior Understanding
Yang, W., Sui, Y., Zhang, Y., & Xia, S. (2025) · International Workshop on Human-Centered Sensing, Modeling, and Intelligent Systems · doi:10.1145/3722570.3726885 · Cited by 3
“Early research in human-centered computing has explored supervised learning approaches to address challenges in Human Activity Recognition (HAR), using Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and deeper models [8, 10, 26, 27].”
Zhang, X., Wang, Y., & Deng, T. (2025) · Scientific Reports · doi:10.1038/s41598-025-32859-1 · Cited by 3
Li, Z., Chen, G., Liu, X., Zhang, L., Sun, Q., Wang, K., Wu, H., & Song, A. (2026) · IEEE Internet of Things Journal · doi:10.1109/jiot.2025.3650552 · Cited by 2
“Guan and Pl¨otz [21] tackled challenges in real-life wearable activity recognition by proposing ensembles of deep LSTM networks, which they demonstrate outperform individual models on standard benchmarks through a modified training procedure.”
Sun, Q., Li, Z., Wang, K., Fu, Z., Chen, G., Zhang, L., Wu, H., & Song, A. (2026) · IEEE Internet of Things Journal · doi:10.1109/jiot.2026.3694938 · Cited by 2
“Pioneering RNN-based deep learning for HAR, Guan and Pl¨otz [6] demonstrated that ensembles of deep LSTMs could overcome real-life challenges and outperform individual LSTM models.”
An unsupervised statistical representation learning method for human activity recognition
Abdi, M. F., BabaAli, B., & Momeni, S. (2024) · Signal Image and Video Processing · doi:10.1007/s11760-024-03374-z · Cited by 3
“[ 27], demonstrated the ben-efits of ensembles of deep LSTM learners with epoch-wise bagging for sample-wise activity recognition.”
Game of LLMs: Discovering Structural Constructs in Activities using Large Language Models
Hiremath, S. K., & Plötz, T. (2024) · ACM International Joint Conference on Pervasive and Ubiquitous Computing (UbiComp) · doi:10.1145/3675094.3678444 · Cited by 4
“Once identified such components are modeled using sequential procedures like HMMs, LSTMs, or RNNs in general [15, 32].”
Self-citationWang, X., Zhang, L., Cheng, D., Tang, Y., Wang, S., Wu, H., & Song, A. (2024) · IEEE Transactions on Instrumentation and Measurement · doi:10.1109/tim.2024.3481549 · Cited by 3
“In terms of precision, recall, accuracy, and F 1 score, we compare our method with several highly competitive algorithms obtained from the previous literature, including [17], [33], [34], [35], [36], [37], [38], and [45].”
MethodMekruksavanich, S., & Jitpattanakul, A. (2025) · Joint International Conference on Digital Arts, Media and Technology with ECTI Northern Section Conference on Electrical, Electronics, Computer and Telecommunications Engineering (ECTI DAMT & NCON) · doi:10.1109/ectidamtncon64748.2025.10962017 · Cited by 2
“Guan and Pl¨otz [29] proposed an ensemble of deep LSTM learners that achieved state-of-the-art performance on several standard datasets.”
Effective Ways to Overcome Classification Limitations for Activities of Daily Livings (ADLs)
Dedabrishvili, M. (2020) · IEEE International Conference on System Analysis & Intelligent Computing (SAIC) · doi:10.1109/saic51296.2020.9239238 · Cited by 2
“The statistical based algorithm Naïve Bayes Classifier and distance-based algorithm K-Nearest Neighbor are often used in prediction problem [6], [7].”
BackgroundMethodPointing at Moving Robots: Detecting Events from Wrist IMU Data
Abbate, G., Gromov, B., Gambardella, L. M., & Giusti, A. (2021) · Proceedings - IEEE International Conference on Robotics and Automation/Proceedings · doi:10.1109/icra48506.2021.9561387 · Cited by 4
“LSTM networks have many successful applications in Human Activity Recognition tasks based on wearable data [17, 18] and can potentially capture complex temporal patterns (and similarities) in the input signals.”
BackgroundAttention-Based Cross-Domain Gesture Recognition Using WiFi Channel State Information
Hong, H., Huang, B., Gu, Y., & Jia, B. (2022) · Lecture notes in computer science · doi:10.1007/978-3-030-95388-1_38 · Cited by 4
Mixing temporal experts for Human Activity Recognition
Roy, D., & Girdzijauskas, Š. (2022) · Swedish Artificial Intelligence Society Workshop (SAIS) · doi:10.1109/sais55783.2022.9833028 · Cited by 2
CASL: Capturing Activity Semantics Through Location Information for Enhanced Activity Recognition
Zhang, X., Shan, C., Zhu, T., Chen, L., Zhou, F., & Ning, H. (2023) · IEEE Transactions on Computational Biology and Bioinformatics · doi:10.1109/tcbb.2023.3238064 · Cited by 2
“We select three feature extraction methods in [39] (LSTM [40], RKHS [41] and DDNN [39]) to compare with heat map methods of CASL, and the results are shown in Figs.”
MethodOrchestrating Development Lifecycle of Machine Learning Based IoT Applications: A Survey.Preprint
Qian, B., Su, J., Wen, Z., Jha, D. N., Li, Y., Guan, Y., Puthal, D., James, P., Yang, R., Zomaya, A. Y., et al. (2019) · arXiv (Cornell University) · Cited by 3
Self-citationChen, X., Guan, Y., Shi, J. Q., Du, X., & Eyre, J. A. (2020) · arXiv (Cornell University) · Cited by 2
Self-citationMultimodal Federated Learning.Preprint
Zhao, Y., Barnaghi, P. M., & Haddadi, H. (2021) · arXiv (Cornell University) · Cited by 4
Sansano‐Sansano, E. (2020) · Tesis Doctorals en Xarxa (Consorci de Serveis Universitaris de Catalunya) · doi:10.6035/14101.2020.180044 · Cited by 4
Human Activity Recognition Dataset for Pedestrians with Mobility Disabilities
Woo, Y., Hwang, S., Oh, S., Kang, M., Lee, S., Kim, J., Cha, J., & Kim, K. K. (2026) · Scientific Data · doi:10.1038/s41597-025-06527-y · Cited by 3
“HAR studies are CNN-based 10 and LSTM-based 53 .”
Li, X. (2022) · Computational Intelligence and Neuroscience · doi:10.1155/2022/1662311 · Cited by 1
Lovanshi, M., Tiwari, V., Singh, A., & Ingle, R. (2024) · IEEE International Conference on ICT in Business Industry & Government (ICTBIG) · doi:10.1109/ictbig64922.2024.10911781 · Cited by 1
“[21] 72.60% Chang et.al.”
InfluentialA Knowledge Graph Framework for Interpretable Video-Based Activity Recognition
Mihailescu, R. (2025) · Proceedings of the AAAI Symposium Series · doi:10.1609/aaaiss.v6i1.36041 · Cited by 1
“Ensembles of deep LSTM learners are explored in (Guan and Pl¨otz 2017) and shown to outperform individual LSTM networks.”
Radio Signal Recognition Using Two-Stage Spatiotemporal Network with Bispectral Analysis
Bai, H., Li, S., Yong Qiang, J. I. A., & Xiao, B. (2025) · Sensors · doi:10.3390/s25175449 · Cited by 1
“The proposed method brings spatiotemporal pattern recognition and multimodal feature fusion into RF identification, drawing on their successful use in radar-RF fusion [24], UAV surveillance [25], and wearable sensor analysis [26].”
CLUES: Cognitive Load Understanding through Experimental Sensing Dataset
Krstevska, A., Goyal, S., Fiorini, L., Bona, F. B. D., Kirilenko, D., Li, M., Trojer, S., Božak, T., Anžur, Z., Luštrek, M., et al. (2026) · Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies · doi:10.1145/3831656 · Cited by 2
“Following established practice in multimodal wearable sensing [21, 58], we considered two broad families.”
FedFitTech: A Baseline in Federated Learning for Fitness Tracking
Oz, Z., Korde, S., Bock, M., & Laerhoven, K. V. (2025) · ACM International Joint Conference on Pervasive and Ubiquitous Computing (UbiComp) · doi:10.1145/3714394.3756332 · Cited by 1
“Over the past decade, Deep Learning (DL) methods have become the dominant approach in inertial-based HAR, consistently outper-forming traditional Machine Learning algorithms [24, 25].”
Transformer-Based Human Activity Recognition Using Wearable Devices
Kumar, A., & Kumar, D. I. (2026) · National Conference on Communications · doi:10.1109/ncc68160.2026.11478937 · Cited by 1
“Unlike conventional CNN–RNN hybrid models [6], [21], the proposed model integrates convolutional layers for efficient local feature extraction with transformer encoder layers [9] for modeling global temporal dependencies.”
Vennela, K., Balaji, B., Chinnaiah, M. C., Lam, S.-K., Janardhan, N., Raju, P. S., Krishna, D. H., Vani, G. D., & Basha, M. (2026) · Sensors · doi:10.3390/s26144644 · Cited by 1
A Real-Time Communication Framework for Distributed Wearable Human Activity Recognition
Rivas-Caicedo, J. L., Saldaña-Aristizábal, L., Niño-Tejada, K., & Patarroyo-Montenegro, J. F. (2026) · Electronics · doi:10.3390/electronics15163714 · Cited by 1
“…short-term memory (CNN–LSTM) architectures have improved the representation of spatial and temporal motion patterns [3–7], while ensemble learning, hybrid models, domain adaptation, smartphone-based methods, and real-time deep architectures have further advanced classification performance [8–13].”
TransfHAR: Self-Supervised Wrist Representations for On-Demand Activity Recognition
Bradshaw, A., Arakawa, R., Liu, X., & Ahuja, K. (2026) · ACM Symposium on User Interface Software and Technology (UIST) · doi:10.1145/3830398.3830553 · Cited by 1
“A separate line of work studies label alignment more directly through few-shot learning and invariant feature methods using limited labeled target data [16, 24, 49, 57].”
Wang, C. (2019) · International Conference on Affective Computing and Intelligent Interaction Workshops and Demos (ACIIW) · doi:10.1109/aciiw.2019.8925091 · Cited by 1
Methods for preprocessing time and distance series data from personal monitoring devices
Wiktorski, T., Bjørkavoll-Bergseth, M. F., & Ørn, S. (2020) · MethodsX · doi:10.1016/j.mex.2020.100959 · Cited by 1
“In contrast, Guan et al. applied LSTM to the actual heart rate data [9] .”
BackgroundKim, H., & Lee, D. (2021) · ACM Symposium on Applied Computing · doi:10.1145/3412841.3441924 · Cited by 2
“LSTM Ensemble : An approach ensembles homogeneous LSTM models [13].”
BackgroundHuman Activity Recognition based on Smartphone Sensors- A Comparative Study
Dhammi, L., & Tewari, P. (2021) · International Conference on Reliability, Infocom Technologies and Optimization (Trends and Future Directions) (ICRITO) · doi:10.1109/icrito51393.2021.9596305 · Cited by 1
Pellatt, L., & Roggen, D. (2022) · Frontiers in Computer Science · doi:10.3389/fcomp.2022.914330 · Cited by 2
A Multiscale Windows Deep Learning Aproach For Sensor-Based Human Activity Recogniton
Duan, F., Zhu, T., Wang, J., & Liu, Z. (2022) · International Conference on Data Intelligence and Security (ICDIS) · doi:10.1109/icdis55630.2022.00074 · Cited by 1
“854 [23] 0.”
MethodResultInfluentialInvestigating Domain-agnostic Performance in Activity Recognition using Accelerometer Data
Hasthanasombat, A., Ghosh, A., Spathis, D., & Mascolo, C. (2022) · ACM International Joint Conference on Pervasive and Ubiquitous Computing (UbiComp) · doi:10.1145/3544793.3560398 · Cited by 2
“The models are often tested on the same dataset used for training [10, 12, 13, 19, 27].”
BackgroundFruit Picker Activity Recognition with Wearable Sensors and Machine Learning
Dabrowski, J. J., & Rahman, A. (2023) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn54540.2023.10191571 · Cited by 2
“Like the CNN, the RNN has been also applied to several HAR problems [28]–[30].”
MethodHAR-STGAN: Human Activity Recognition-Spatial Transformer network for Generative Adversarial Network
Nian, A., Huang, S., Wang, F., & Zhao, Y. (2024) · International Conference on Electronic Technology and Information Science (ICETIS) · doi:10.1109/icetis61828.2024.10593790 · Cited by 1
“These advanced algorithms possess the capability to leverage information from the entire dataset, but often require meticulous tuning to achieve optimal performance [6-8].”
BackgroundMethodHwang, S., Leng, Z., Oh, S., Kim, K. K., & Plötz, T. (2024) · ACM International Symposium on Wearable Computers (ISWC) · doi:10.1145/3675095.3676617 · Cited by 1
“Future research should aim to develop a model that capitalizes on the strengths of both datasets to further enhance wTMD performance (e.g., through Ensmebles [7]).”
Self-citationWi-Fi sensing gesture control algorithm based on semi-supervised generative adversarial network
Wang, C., Ding, Y., Zhou, M. R., & Tang, L. (2024) · PeerJ Computer Science · doi:10.7717/peerj-cs.2402 · Cited by 1
“The method based on wearable sensors has a limited perception range and poor user-friendliness ( Guan & Plötz, 2017 ; Yatani & Truong, 2012 ).”
ATTENTION-ENHANCED CNN-LSTM MODELS FOR SENSOR-BASED HUMAN ACTIVITY RECOGNITION: A COMPARATIVE STUDY
(2025) · International Journal of Mechatronics and Applied Mechanics · doi:10.17683/ijomam/issue20.23 · Cited by 5
Zeng, M., Gao, H., Yu, T., Mengshoel, O. J., Helge Langseth, A., Lane, I. R., & Liu, X. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1810.04038 · Cited by 7
“[8] used ensemble LSTM model to capture diverse data during the training and significantly improved the recognition performance and robustness.”
BackgroundMethodResultInfluentialMeta-Decomposition: Dynamic Segmentation Approach Selection in IoT-based Activity RecognitionPreprint
Modaresi, S. M. R., Osmani, A., Razzazi, M., & Chibani, A. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2404.11742 · Cited by 1
“Guan and Plötz [Guan and Plötz, 2017] claim that the deep learning method can be insensitive to the window size in HAR.”
BackgroundWithdrawn: Leveraging LDA Feature Extraction to Augment Human Activity Recognition AccuracyPreprint
Vazan, M., Sharifi, E., Farahani, H., & Madadi, S. (2024) · Preprints.org · doi:10.20944/preprints202406.1528.v1 · Cited by 1
Mirzaseyedi, M. (2024) · Thesis Commons (OSF Preprints) · doi:10.31237/osf.io/eamsy · Cited by 1
Efficient Online Continual Learning in Sensor-Based Human Activity RecognitionPreprint
Zhang, Y., Leite, C. S., & Xiao, Y. (2026) · SSRN Electronic Journal · doi:10.2139/ssrn.7161268 · Cited by 1
Nguyen, D.-V. (2018) · theses.fr (ABES) · doi:10.70675/1a8f808bz7418z43c4z84e4z9a688e16eae6 · Cited by 3
Gao, X., Zhang, H., Zhang, Z., Ruan, J., Liu, T., & Fu, Y. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2604.17065 · Cited by 1
“Deep learning models, including CNNs [10, 18, 22], RNNs [11, 16], and LSTMs [2, 8], are commonly employed for feature extraction and classification of these signals.”
Burzer, M., King, T., Riedel, T., Beigl, M., & Röddiger, T. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.13194 · Cited by 1
“• Guan-LSTM [37] is included in the benchmark for coverage of this architectural line, but under our standardized training pipeline it is trained as a regular single model rather than as an ensemble.”
Haresamudram, H., Tang, C. I., Suh, S., Lukowicz, P., & Ploetz, T. (2024) · arXiv · Cited by 31
“…[137] were introduced specifically eliminating the explicit feature design phase but rather utilizing proven feature learners such as convolutional blocks into the model architectures, complemented by explicit sequential modeling parts in form of LSTM blocks [83] or later ensembles thereof [65].”
Self-citationPhoneotypic modeling of human behaviors and propensities
Bati, G. F. (2019) · doi:10.7282/t3-mndr-a738
Sensing-HH: A Deep Hybrid Attention Model for Footwear Recognition
Yao, Y., Wen, Y., & Wang, J. (2020) · Electronics · doi:10.3390/electronics9091552
“Inspired by the deep neural networks, some very recent works employ them to motion sensor-based recognition, such as Convolution Neural Network (CNN) [3,5,23], which are competent in capturing the local characteristics of multi-channel signals; Recurrent Neural Network (RNN) [24], and its variant, LSTM units [1,25], w…”
MethodAbe, H., Hino, T., Sugihara, M., Ikeya, H., & Shimosaka, M. (2020) · IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) · doi:10.1109/iros45743.2020.9341500
“From this characteristic, LSTM has been frequently applied to activityrecognition researches in recent years [8], [9], [7].”
InfluentialSensor Fusion in Human Activity Recognition and Occupancy Detection
Zimmermann, L. (2020) · OPUS FAU (Kooperativer Bibliotheksverbund Berlin-Brandenburg (KOBV), on behalf of the Universitätsbibliothek Erlangen-Nürnberg) · Cited by 1
Hitting the Success Notes of Deep Learning
Aggarwal, S., Singh, N., & Mishra, K. K. (2021) · Wiley eBooks · doi:10.1002/9781119769262.ch5
Human Gesture Recognition with Wi-Fi
Yang, Z., Qian, K., Wu, C., & Zhang, Y. (2021) · Springer eBooks · doi:10.1007/978-981-16-5658-3_9
Hierarchical classification and visualization with multiple feature ranking criteria
Wang, Q., Wang, X., Zhou, Z., Sheng, D., & Sheng, D. (2021) · International Congress on Image and Signal Processing (CISP) · doi:10.1109/cisp-bmei53629.2021.9624378
RECOGNITION OF HUMAN ACTIONS IN VIDEO SEQUENCES USING THE LSTM NETWORK
Teplyakova, A. R., & Старков, С. О. (2021) · Vestnik komp iuternykh i informatsionnykh tekhnologii · doi:10.14489/vkit.2021.08.pp.012-021
An Advanced Attention-Enhanced Hybrid Deep Learning Model for WBSNs’ Gait Pattern RecognitionPreprint
Wu, J., Tan, Q., & Wu, X. (2022) · Research Square · doi:10.21203/rs.3.rs-1251973/v1
“These algorithms mainly include convolutional neural network (CNN), dilated convolutional neural network (DCNN), recurrent neural network (RNN) and long short-term memory(LSTM) [12-15].”
BackgroundGonzalez, D. R., & Qu, Y. (2022) · European Journal of Electrical Engineering and Computer Science · doi:10.24018/ejece.2022.6.2.433
“As expressed by Guan and Plötz , unconstrained n-fold cross-validation is avoided [5].”
BackgroundMethodInfluentialPhát hiện hoạt động bất thường của người bằng mạng học sâu tích chập kết hợp mạng bộ nhớ dài ngắn
Tuấn, L. N., Pham, C. H., & Văn, T. N. (2020) · Journal on Information Technologies & Communications · doi:10.32913/mic-ict-research-vn.v2020.n1.925
“Trước đây, phương pháp tiếp cận phát hiện hoạt động bất thường chủ yếu dựa trên các mô hình học máy trong đó học có giám sát [12] được sử dụng phổ biến.”
BackgroundComputer-Cognition Interfaces: Sensing and influencing mental processes with computer interactionPreprint
Mottelson, A. (2019) · Thesis Commons (OSF Preprints) · doi:10.31237/osf.io/hg378
Ageing@Work Worker Smart Tracking infrastructure
Alexandra, A., Chatzikonstantinou, I., Giakoumis, D., Mastoras, R.-E., Paliokas, I., Evangelos, P., & Andreas, T. (2021) · Zenodo (CERN European Organization for Nuclear Research) · doi:10.5281/zenodo.5162899
A Transferable Framework for Few-Shot Human Activity Recognition
Zhang, Y., Wang, H., Tian, F., & Tian, Z. (2022) · Communications in computer and information science · doi:10.1007/978-3-031-06767-9_50
Sahoo, G. S. P. A. K., & Qu, Y. (2024) · European Journal of Electrical Engineering and Computer Science · doi:10.24018/ejece.2024.8.1.600
A review on devices and learning techniques in domestic intelligent environment
Ye, J., Wang, M., Zhong, J., & Jiang, H. (2024) · Journal of Ambient Intelligence and Humanized Computing · doi:10.1007/s12652-024-04759-1
Gonzalez, D. R., & Qu, Y. (2023) · Congress in Computer Science, Computer Engineering, and Applied Computing (CSCE) · doi:10.1109/csce60160.2023.00084 · Cited by 1
“Down-sampling has been used by Guan and Plötz and Lv et al. [2][10].”
BackgroundMethodInfluentialOshima, C., Takatsu, T., & Nakayama, K. (2024) · Lecture notes in computer science · doi:10.1007/978-3-031-60125-5_8
Beyond Thresholds: A General Approach to Sensor Selection for Practical Deep Learning-based HAR
Cooper, G., & Marculescu, R. (2024) · IEEE/ACM International Conference on Internet-of-Things Design and Implementation (IoTDI) · doi:10.1109/iotdi61053.2024.00005
“7, 8, 9 for examples.”
BackgroundMirzaseyedi, M., Khatibi, T., & Dehghanian, A. (2024) · doi:10.36227/techrxiv.172296262.28312060/v1
A Survey on Deep Learning Based Human Activity Recognition System
Thomas, A. L., & Judith, J. E. (2024) · Communications in computer and information science · doi:10.1007/978-3-031-73065-8_10
Real-Time Human Activity Recognition for Elderly VR Training with Body Area Networks
Fan, Y.-C., & Wen, C.-Y. (2024) · Studies in computational intelligence · doi:10.1007/978-3-031-71821-2_8
Human-AI Collaboration In Rehabilitation Assessment: A New Clinical Paradigm
Sabr, A. R. H., Abdulaziz, X. I., Saedy, A. S., Shakir, K. H., Mejbil, H. G., Al–Fatlawy, R. R., & Mohammed, A. H. (2024) · International Conference on IoT, Communication and Automation Technology (ICICAT) · doi:10.1109/icicat62666.2024.10923063
Multi-head Temporal Learning for Human Activity Recognition Using Smartphone
Wei, S., Han, P. Y., How, K. W., & Yin, O. S. (2025) · Smart innovation, systems and technologies · doi:10.1007/978-981-97-7786-0_29
Liu, X., Zhang, L., Huang, W., Xiong, D., Fu, Z., Wu, H., & Song, A. (2025) · IEEE Transactions on Instrumentation and Measurement · doi:10.1109/tim.2025.3566808
“Guan and Plötz [13] have proposed to ensemble multiple deep long short-term memory (LSTM) learners, which significantly outperform individual LSTM networks, thus pushing the state-of-the-art in wearable-based activity recognition.”
InfluentialOn the Utilization of Sensor-Based Machine Learning Approaches for Human Activity Recognition
Hamid, S., Javed, T., Khan, A., & Iqbal, S. (2025) · Journal of Engineering and Applied Sciences , University of Engineering and Technology, Peshawar · doi:10.17582/journal.jeas/44.13.26
HAPX-CLIP: Human Activity Recognition with Visual Sequences and Language Prompts
Nguyen, D.-B., Nguyen, T.-K., Le, T.-H., & Tran, T.-H. (2025) · International Conference on Multimedia Analysis and Pattern Recognition (MAPR) · doi:10.1109/mapr67746.2025.11133777
“These models, as demonstrated by Guan et al. [14], are capable of capturing activity evolution over time, but often suffer from limited ability to model long-range sequences due to vanishing gradient issues.”
Human Activity Recognition Using Machine Learning
Tiwari, M., & Gupta, P. (2025) · Communications in computer and information science · doi:10.1007/978-3-032-00793-3_12
Kaithal, P. K., Sengar, A. S., Chourey, A., Gupta, C., Pandey, A., & Gangrade, S. (2025) · World Skills Conference on Universal Data Analytics and Sciences (WorldSUAS) · doi:10.1109/worldsuas66815.2025.11199300
Anikwe, C. V., Nweke, H. F., & Omeje, K. N. (2025) · African Journal of Advances in Science and Technology Research · doi:10.62154/ajastr.2025.021.01012
Rad, N. M. (2019) · Unitn-eprints PhD (University of Trento) · Cited by 1
Human action recognition in video data based on deep learning methods
Shaker, S. H., Mahdi, A. F., Hasan, R. F., & Azeez, R. A. (2025) · AIP conference proceedings · doi:10.1063/5.0301806
Leveraging Simulated Sensor Data and Foundation Models for Human Motion Understanding
Ray, L. S. S. (2025) · ACM International Joint Conference on Pervasive and Ubiquitous Computing (UbiComp) · doi:10.1145/3714394.3750552
“However, developing robust HMU models remains challenging due to their dependence on large, diverse, and high-quality sensor datasets that capture variations in user behavior, sensor placement, clothing, and environmental factors [10].”
Kurniawan, A., Arifin, S., Siregar, A. M., Ariyanto, M., Mursalim, M. K. N., & Hakim, D. L. (2025) · SSRN Electronic Journal · doi:10.20944/preprints202512.2300.v1
Metadata
- Authors
- Guan, Y., & Plötz, T.
- Affiliations
- Newcastle University (GB)
- Georgia Institute of Technology (US)
- ORCID
- Guan, Yu
- Venue
- Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
- Published
- 2017-06-30
- Volume
- 1
- Issue
- 2
- Pages
- 1-28
- ISSN
- 2474-9567 (ISSN-L)
- OpenAlex
- W2604630936
- Cited by
- 511
- References
- 65
- Citations by year
- 2026: 33 · 2025: 35 · 2024: 59 · 2023: 60 · 2022: 70 · 2021: 82 · 2020: 87 · 2019: 46 · 2018: 33 · 2017: 6
- FWCI
- 19.53
- Type
- journal article
- Language
- en
- Open access
- green
- arxiv.org
- Locations
- Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Funders
- Engineering and Physical Sciences Research Council (EP/M020576/1)
- Topics
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Context-Aware Activity Recognition Systems
- Physical Sciences › Computer Science › Human-Computer Interaction › Hand Gesture Recognition Systems
- Physical Sciences › Engineering › Biomedical Engineering › Gait Recognition and Analysis
- Keywords
- human activity recognition, wearable sensors, deep learning, ensemble classifiers, class imbalance, data quality, training protocols, PAMAP2, Opportunity dataset
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.
A language model drew the claims from the paper on 2026-10-05, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-05 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
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.