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 301–400 of 539
A Method of Self-Supervised Denoising and Classification for Sensor-Based Human Activity Recognition
Lu, L., & Deng, T. (2023) · IEEE Sensors Journal · doi:10.1109/jsen.2023.3323314 · Cited by 10
“Long short-term memory (LSTM) [17] and gated recurrent unit (GRU) [18] are RNN variants used to solve gradient disappearance and gradient explosion problems.”
Incept_LSTM : Accession for human activity concession in automatic surveillance
Girdhar, P., Johri, P., & Virmani, D. (2020) · Journal of Discrete Mathematical Sciences and Cryptography · doi:10.1080/09720529.2020.1804132 · Cited by 10
“In [17] Authors proposed multimodal system for concurrent activity recognition.”
BackgroundTemporal Convolutional Network with Wavelet Transform for Fall Detection
Lu, X., Ling, Y., & Liu, S. (2022) · Journal of Sensors · doi:10.1155/2022/7267099 · Cited by 10
“Models include support vector machine (SVM) [14], Google’s deep neural network (DNN) [15], convolutional neural network (CNN) [16–19], long short-term memory network (LSTM) [20, 21], and recurrent neural network (RNN) [22–24].”
An Enhanced Deep Learning Approach for Smartphone-Based Human Activity Recognition in IoHT
Soni, V., Jaiswal, S., Semwal, V. B., Roy, B., Choubey, D. K., & Mallick, D. K. (2023) · Lecture notes in electrical engineering · doi:10.1007/978-981-19-5868-7_37 · Cited by 12
Unsupervised Domain Adaptation for WiFi Gesture Recognition
Zhang, B., Zhang, D., Hu, Y., & Chen, Y. (2023) · IEEE Wireless Communications and Networking Conference · doi:10.1109/wcnc55385.2023.10118941 · Cited by 10
“…which could be applied to smart home, smart driving and virtual reality, etc. Traditional approaches adopt cameras [1], wearable devices and phones [2], [3] as the sensing front-end, which however suffer from inherent drawbacks including privacy leakage, inconvenience, and limited sensing range.”
BackgroundInvestigating (re)current state-of-the-art in human activity recognition datasets
Bock, M., Hoelzemann, A., Moeller, M., & Laerhoven, K. V. (2022) · Frontiers in Computer Science · doi:10.3389/fcomp.2022.924954 · Cited by 10
“, 2016) and assessed the applicability in real-world scenarios (Guan and Plötz, 2017).”
BackgroundAhmad, N., & Leung, H.-F. (2024) · Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies · doi:10.1145/3643511 · Cited by 13
“…variations of recurrent neural networks (RNN) are investigated for HAR, e.g., forward, and bidirectional long short-term memory (LSTM) [24] and ensemble method of various LSTM models [22] to learn the activity sequence’s temporal dependencies and capture the whole sequence’s global representation.”
BackgroundMethodResultInfluentialSpectrum-Guided Adversarial Disparity Learning
Liu, Z., Yao, L., Bai, L., Wang, X., & Wang, C. (2020) · ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD) · doi:10.1145/3394486.3403054 · Cited by 9
“…Neural Network • Bi-LSTM-S [10]: a bi-directional Long Short-term Memory • ConvLSTM [18]: a hybrid model that combines both CNN and • En-LSTM [9]: an LSTM-based method that combines multiple • AttConvLSTM [17]: a hybrid model that applies attention layer • AAE [16]: a benchmark of…”
Confidence-Calibrated Human Activity Recognition
Roy, D., Girdzijauskas, Š., & Socolovschi, S. (2021) · Sensors · doi:10.3390/s21196566 · Cited by 9
“Unlike [6], the window sizes are selected in a non-random fashion.”
BackgroundMethodResultInfluentialViSig: Automatic Interpretation of Visual Body Signals Using On-Body Sensors
Cao, Y., Dhekne, A., & Ammar, M. (2023) · Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies · doi:10.1145/3580797 · Cited by 14
“[29] improves model robustness by combining multiple LSTM learners.”
BackgroundMethodResultHuman Activity Recognition via Temporal Fusion Contrastive Learning
Kim, I., Lim, J., & Lee, J. (2024) · IEEE Access · doi:10.1109/access.2024.3357143 · Cited by 10
“Most existing HAR studies are based on supervised learning [15], [16], [17], [18], [19].”
MethodCompressed Deep Learning Model For Human Activity Recognition
k, A. A. O., Semwal, V. B., & Soni, V. (2024) · IEEE International Students' Conference on Electrical, Electronics and Computer Science (SCEECS) · doi:10.1109/sceecs61402.2024.10482048 · Cited by 11
“Few studies have addressed these factors, with some like Lane et al.’s [7], [8] approach being unreproducible due to inaccessible datasets.”
BackgroundMulti-context embedding based personalized place semantics recognition
Chen, L., Han, M., Shi, H., & Liu, X. (2020) · Information Processing & Management · doi:10.1016/j.ipm.2020.102416 · Cited by 9
Adversary Helps: Gradient-based Device-Free Domain-Independent Gesture RecognitionPreprint
Liu, J., Han, J., Lin, F., & Ren, K. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2004.03961 · Cited by 9
“Wearable sensor-based approaches [5], [6], though are light and visual information-needless, they face the flaws of limit of sensing range and relatively bad user-friendness.”
BackgroundTFC: Time–frequency contrasting network for wearable-based human activity recognition
Huang, Z., Deng, J., Wang, S., Tang, C., & Xiao, S. (2025) · Knowledge-Based Systems · doi:10.1016/j.knosys.2025.113373 · Cited by 9
Arı hastalıklarının hibrit bir derin öğrenme yöntemi ile tespiti
Metlek, S., & Kayaalp, K. (2021) · Gazi Üniversitesi Mühendislik-Mimarlık Fakültesi Dergisi · doi:10.17341/gazimmfd.749443 · Cited by 12
Huang, Q., Xie, W., Li, C., Wang, Y., & Liu, Y. (2023) · Applied Sciences · doi:10.3390/app131910560 · Cited by 9
Harnessing Deep Learning for Activity Recognition in Seniors' Daily Routines with Wearable Sensors
Mekruksavanich, S., Phaphan, W., & Jitpattanakul, A. (2024) · International Conference on Telecommunications and Signal Processing (TSP) · doi:10.1109/tsp63128.2024.10605946 · Cited by 8
“Hybrid architectures, like EnsembleLSTM by Guan and Pl¨otz [29], combine LSTM networks for improved accuracy.”
MethodA hybrid TCN-GRU model for classifying human activities using smartphone inertial signals
Sekaran, S. R., Pang, Y. H., You, L. Z., & Yin, O. S. (2024) · PLoS ONE · doi:10.1371/journal.pone.0304655 · Cited by 9
“Since Recurrent Neural Networks (RNN) are susceptible to short-term memory problems where the gradient tends to either explode or vanish during training [30], less adoption has been done in human gait recognition.”
BackgroundVisible Light Based Occupancy Inference Using Ensemble Learning
Hao, J., Yuan, X., Yang, Y., Wang, R., Zhuang, Y., & Luo, J. (2018) · IEEE Access · doi:10.1109/access.2018.2809612 · Cited by 8
“Just to name a few, we can employ sophisticated bagging method for LSTM [36], orientation ordering [35] to accelerate the pruning, or optimization based method to refine the pruning and integration [37].”
MethodKalabakov, S., Gjoreski, M., Gjoreski, H., & Gams, M. (2021) · Informatica · doi:10.31449/inf.v45i2.3648 · Cited by 8
CogAx: Early Assessment of Cognitive and Functional Impairment from Accelerometry
Ramamurthy, S. R., Chatterjee, S., Galik, E. M., Gangopadhyay, A., Roy, N., Mitra, B., & Chakraborty, S. (2022) · IEEE International Conference on Pervasive Computing and Communications · doi:10.1109/percom53586.2022.9762401 · Cited by 11
The Human Continuity Activity Semisupervised Recognizing Model for Multiview IoT Network
Yuan, R., & Wang, J. P. (2023) · IEEE Internet of Things Journal · doi:10.1109/jiot.2023.3234053 · Cited by 8
“In addition, because the time-series sensor data may contain the temporal characteristics of the activity, there are also some studies using recurrent neural networks (RNNs) with long short-term memory [29] to extract the temporal features of original data.”
BackgroundFine-Grained Intoxicated Gait Classification Using a Bilinear CNN
Li, R., Agu, E., Sarwar, A., Grimone, K., Herman, D. S., Abrantes, A. M., & Stein, M. D. (2023) · IEEE Sensors Journal · doi:10.1109/jsen.2023.3248868 · Cited by 8
“In our experiments, the walking speed is calculated by dividing the distance covered by subjects based on the walking protocol d) Gait data normalization: As motivated by prior work, our gait dataset should be segmented, standardized, and normalized [21], [45], [47].”
MethodALAE-TAE-CutMix+: Beyond the State-of-the-Art for Human Activity Recognition Using Wearable Sensors
Ahmad, N., & Leung, H.-F. (2023) · IEEE International Conference on Pervasive Computing and Communications · doi:10.1109/percom56429.2023.10099138 · Cited by 13
“For a fair comparison, we use the same evaluation approach (e.g., hold-out), performance metrics, and standard training and testing partitions used in our state-of-the-art studies [1], [10], [11], [16].”
BackgroundMethodInfluentialt-SNE Manifold Learning Based Visualization: A Human Activity Recognition Approach
Ramesh, D., MadhukarRao, G., Khurana, H., & Edla, D. R. (2020) · Lecture notes on data engineering and communications technologies · doi:10.1007/978-981-15-0978-0_3 · Cited by 8
Non-Linear Chaotic Features-Based Human Activity Recognition
Tu, P., Li, J., Wang, H., Cao, T., & Wang, K. (2021) · Electronics · doi:10.3390/electronics10020111 · Cited by 8
“Especially for feature extraction module, the existing researches are roughly divided into two categories: time and frequency domain features [6] and automatically extracted features based on deep learning [7,8].”
MethodFollower: A Novel Self-Deployable Action Recognition Framework
Yang, X., Liu, D., Liu, J., Yan, F., Chen, P., & Niu, Q. (2021) · Sensors · doi:10.3390/s21030950 · Cited by 9
“Action recognition approaches mainly include the following three categories: based on wearable sensors [1], based on wireless signals [2], and based on vision [3–5].”
BackgroundInfluentialOptimal sensor channel selection for resource-efficient deep activity recognition
Leite, C. F. S., & Xiao, Y. (2021) · ACM/IEEE International Conference on Information Processing in Sensor Networks (IPSN) · doi:10.1145/3412382.3458278 · Cited by 13
“Feature Ours Exhaustive search [1, 8, 9, 30] Sensor channel selection [5, 15, 17, 29] Standard DL methods [10, 11, 19, 20, 28]”
InfluentialDevice-free cross location activity recognition via semi-supervised deep learning
Zhou, R., Gong, Z., Tang, K. W., Zhou, B., & Cheng, Y. (2022) · Neural Computing and Applications · doi:10.1007/s00521-022-07085-9 · Cited by 9
“Conventional approaches to human activity recognition are mainly based on vision [7, 29] and wearables [9, 21].”
BackgroundEvaluation of Transfer Learning for Human Activity Recognition Among Different Datasets
Islam, M. S., Okita, T., & Inoue, S. (2019) · IEEE Intl Conf on Dependable, Autonomic and Secure Computing, Intl Conf on Pervasive Intelligence and Computing, Intl Conf on Cloud and Big Data Computing, Intl Conf on Cyber Science and Technology Congress (DASC/PiCom/CBDCom/CyberSciTech) · doi:10.1109/dasc/picom/cbdcom/cyberscitech.2019.00155 · Cited by 11
“Recursive Neural Network (RNN) model [17], Long Short Term Memory (LSTM) have been used in human activity recognition and have achieved very good accuracy.”
MethodDesigning Compact Features for Remote Stroke Rehabilitation Monitoring using Wearable Accelerometers
Chen, X., Guan, Y., Shi, J. Q., Du, X.-L., & Eyre, J. (2020) · arXiv · doi:10.48550/arxiv.2009.08798 · Cited by 8
“,[11][30]), the annotation here is inadequate.”
BackgroundSelf-citationDeep Learning Algorithms for Human Activity Recognition in Manual Material Handling Tasks
Bassani, G., Avizzano, C. A., & Filippeschi, A. (2025) · Sensors · doi:10.3390/s25216705 · Cited by 7
“Most of them attempt fixed window sizes [47], while others use random length and position frames [58].”
A Tailored Textile Sensor-based Wrap for Shoulder Complex Angles Monitoring
Zhang, R., Hu, Y., He, Y., Fang, S., Yu, B., Sun, X., & Wang, Q. (2023) · ACM International Joint Conference on Pervasive and Ubiquitous Computing (UbiComp) · doi:10.1145/3594739.3610707 · Cited by 7
Indoor Abnormal Behavior Detection for the Elderly: A Review
Gu, T., & Tang, M. (2025) · Sensors · doi:10.3390/s25113313 · Cited by 10
“Guan et al. [53] proposed an ensemble learning method based on LSTM networks to train multiple LSTM models by randomly selecting a subset of the training data, which are then integrated for final activity identification.”
Bati, G. F., & Singh, V. K. (2020) · Computers · doi:10.3390/computers10010003 · Cited by 8
“There has been a rich array of recent work on modeling human activities using sensors and deep learning [13,14,49,50].”
BackgroundLeveraging a Smartwatch for Activity Recognition in Salat
Eliza, I. J., Al-Nabhan, N., Noor, J., Rahaman, M., & Islam, A. B. M. A. A. (2023) · IEEE Access · doi:10.1109/access.2023.3311261 · Cited by 9
“have taken a dominant role in the HAR domain [94], [95].”
BackgroundLi, Y., Wu, J., Li, W., & Fang, A. (2024) · IEEE Transactions on Instrumentation and Measurement · doi:10.1109/tim.2024.3366277 · Cited by 10
“We also introduce some works on handling imbalance for evaluation, including SMOTE and CS-SVM [17], Kim et al. [19], and EnLSTM [20].”
BackgroundMethodRobust Suspicious Action Recognition Approach Using Pose Descriptor
Ahmed, W., Yousaf, M. H., & Yasin, A. (2021) · Mathematical Problems in Engineering · doi:10.1155/2021/2449603 · Cited by 7
“,e runtime complexity comprises two major parts: one CNN processing time with complexity isO (1), constant with a variable number of people; second, multiperson analyzing time, which takes the complexity of O (n2), where n represents the number of actors present in the video.”
InfluentialPellatt, L., & Roggen, D. (2021) · International Symposium on Wearable Computers · doi:10.1145/3460421.3478813 · Cited by 7
“significantly worse than reported in the literature [17, 18], and thus we were not able to show an improvement over the state-of-the-art F1 score.”
MethodResultADL-GAN: Data Augmentation to Improve In-the-Wild ADL Recognition Using GANs
Ditthapron, A., Lammert, A. C., & Agu, E. (2022) · IEEE Access · doi:10.1109/access.2023.3271409 · Cited by 7
“While early ANNs for ADL recognition also analyzed handcrafted sensor features [2], [5], more recent work [4], [6] analyzed raw data directly using powerful ANNs such as Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTMs), which obviate the need for feature engineering step that…”
BackgroundMethodSensor-based human activity recognition using fuzzified deep CNN architecture with λmax method
Gomathi, V., Kalaiselvi, S., & Thamarai Selvi, D. (2022) · Sensor Review · doi:10.1108/sr-06-2021-0195 · Cited by 9
Wang, C., Olugbade, T., Mathur, A., Williams, A. C. D. C., Lane, N. D., & Bianchi‐Berthouze, N. (2019) · arXiv (Cornell University) · Cited by 7
Adversarial Transferability in Wearable Sensor SystemsPreprint
Sah, R. K., & Ghasemzadeh, H. G. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2003.07982 · Cited by 8
Rey, V. F., Rezende, P. M. B., Zhou, B., Suh, S., & Lukowicz, P. (2025) · Expert Systems with Applications · doi:10.1016/j.eswa.2025.129288 · Cited by 14
Optimizing Accuracy-Efficiency Trade-Offs of On-Device Activity Inference With Star Operation
Chen, G., Fu, Z., Sha, Y., Xiong, D., Zhang, L., Wu, H., & Song, A. (2025) · IEEE Journal of Biomedical and Health Informatics · doi:10.1109/jbhi.2025.3619549 · Cited by 6
“However, such convolution-based or hybrid models still struggle to model long-range temporal dependence, preventing further performance gain [22].”
IMUZero: Zero-Shot Human Activity Recognition by Language-Based Cross Modality Fusion
Su, J., Ge, F., Wen, Z., Li, T., Bai, Y., Zhou, Y., & Zhang, X. (2025) · Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies · doi:10.1145/3770669 · Cited by 8
“Numerous works have investigated the CNN variants [11, 13, 26, 27, 34, 49] by integrating with specific novel network architecture (e.g., Long-Short Term Memory(LSTM) [14], Attention Network [26], etc) to enhance the activity pattern extraction ability.”
Zhang, J. (2025) · Journal of Technology Informatics and Engineering · doi:10.51903/jtie.v4i1.524 · Cited by 6
Gonçalves, H. R., Martins, L., Ribeiro, N. F., Abreu, L., Rodrigues, A. M., & Santos, C. P. (2025) · Engineering applications of artificial intelligence · doi:10.1016/j.engappai.2025.110706 · Cited by 6
McDaniel, C., & Quinn, S. P. (2018) · Proceedings of the Python in Science Conferences · doi:10.25080/majora-4af1f417-005 · Cited by 7
Capturing Interaction Quality in Long Duration (Simulated) Space Missions With Wearables
Gedik, E., Olenick, J. S., Chang, C.-H. D., Kozlowski, S. W. J., & Hung, H. (2022) · IEEE Transactions on Affective Computing · doi:10.1109/taffc.2022.3176967 · Cited by 6
Sensor-Based Gait Analysis for Parkinson’s Disease Prediction
Sathya Bama, B., & Jinila, Y. B. (2023) · Intelligent Automation & Soft Computing · doi:10.32604/iasc.2023.028481 · Cited by 6
Detecting Hand Hygienic Behaviors In-the-Wild Using a Microphone and Motion Sensor on a Smartwatch
Zhuang, H., Xu, L., Nishiyama, Y., & Sezaki, K. (2023) · Lecture notes in computer science · doi:10.1007/978-3-031-34609-5_34 · Cited by 7
SleepNet: Attention-Enhanced Robust Sleep Prediction using Dynamic Social Networks
Khalid, M., Klerman, E. B., McHill, A. W., Phillips, A. J. K., & Sano, A. (2024) · Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies · doi:10.1145/3643508 · Cited by 8
“This architecture is identical to the proposed method except that the graph attention network is replaced by CNN. • LSTM only: In order to observe the improvement provided by graph integration, we also evaluate the model with LSTM layers [25] only and no GAN.”
MethodWeighted voting ensemble of hybrid CNN-LSTM Models for vision-based human activity recognition
Aggarwal, S., Bhola, G., & Vishwakarma, D. K. (2024) · Multimedia Tools and Applications · doi:10.1007/s11042-024-19582-1 · Cited by 7
Exploring the Impact of the NULL Class on In-the-Wild Human Activity Recognition
Cherian, J., Ray, S., Taele, P., Koh, J. I., & Hammond, T. (2024) · Sensors · doi:10.3390/s24123898 · Cited by 6
“The former is a deep learning network comprised of convolutional and LSTM recurrent layers and has been used as a benchmark in several recent studies [21,84].”
MethodDeep Learning for Sensor-Based Activity Recognition: Recent Trends
Ahad, M. A. R., Antar, A. D., & Ahmed, M. (2020) · Intelligent systems reference library · doi:10.1007/978-3-030-51379-5_9 · Cited by 7
A Hybrid Deep Neural Networks for Sensor-based Human Activity Recognition
Wang, S., & Zhu, X. (2020) · International Conference on Advanced Computational Intelligence (ICACI) · doi:10.1109/icaci49185.2020.9177818 · Cited by 6
“Yu and Thomas [9] employed ensembles of deep long and short term memory (LSTM) learners to human activity recognition.”
BackgroundGravity Control-Based Data Augmentation Technique for Improving VR User Activity Recognition
Han, D., Lee, C., & Kang, H. (2021) · Symmetry · doi:10.3390/sym13050845 · Cited by 7
DeepFit: deep learning based fitness center equipment use modeling and prediction
Kulkarni, A., Seetharam, A., & Ramesh, A. (2019) · EAI International Conference on Mobile and Ubiquitous Systems: Computing, Networking and Services (MobiQuitous) · doi:10.1145/3360774.3360803 · Cited by 6
“In comparison, the authors in [10] combine multiple LSTM learners into ensemble classifiers for activity recognition.”
BackgroundMethodCombined deep centralized coordinate learning and hybrid loss for human activity recognition
Bourjandi, M., Yadollahzadeh‐Tabari, M., & Golsorkhtabaramiri, M. (2022) · Concurrency and Computation Practice and Experience · doi:10.1002/cpe.6870 · Cited by 6
T-WaveNet: Tree-Structured Wavelet Neural Network for Sensor-Based Time Series AnalysisPreprint
Liu, M., Zeng, A., Lai, Q., & Xu, Q. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2012.05456 · Cited by 6
“[Guan and Pl¨otz, 2017; Yao et al. , 2017; Chen et al. , 2016; Ord´o˜nez and Roggen, 2016] combine CNNs with Recurrent neural networks (RNNs) to extract local and long-term temporal dependency features.”
BackgroundColloSSL: Collaborative Self-Supervised Learning for Human Activity RecognitionPreprint
Jain, Y., Tang, C. I., Min, C., Kawsar, F., & Mathur, A. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2202.00758 · Cited by 7
“Later, deep convolutional and recurrent networks [18, 19, 64] were shown to be effective in learning feature extractors from labeled data.”
BackgroundDhar, J., Rana, K., Goyal, P., Alavi, A., Rana, R. K., Vo, Q. B., Mishra, S., & Mistry, S. (2024) · Applied Soft Computing · doi:10.1016/j.asoc.2024.112550 · Cited by 5
Liu, X., Zhang, L., Huang, W., Cheng, D., Wu, H., & Song, A. (2025) · Applied Soft Computing · doi:10.1016/j.asoc.2025.113861 · Cited by 5
Personalised Gait Recognition for People with Neurological Conditions
Ingelse, L., Branco, D., Gjoreski, H., Guerreiro, T., Bouça‐Machado, R., Ferreira, J. J., & The CNS Physiotherapy Study Group (2022) · Sensors · doi:10.3390/s22113980 · Cited by 5
Rahman, A., Nahid, N., Schuller, B. W., & Ahad, M. A. R. (2024) · Scientific Reports · doi:10.1038/s41598-024-81228-x · Cited by 6
Evaluating the Effectiveness of Plantar Pressure Sensors for Fall Detection in Sloped Surfaces
Mahmud, T., Kayastha, R., Kisi, K., Ngu, A. H. H., & Alamgeer, S. (2025) · Electronics · doi:10.3390/electronics14153003 · Cited by 5
Robust Human Activity Recognition Using a Transformer-Based Model for Aging Society
Liang, Q., Jitpattanakul, A., & Mekruksavanich, S. (2025) · International Conference on Big Data Analytics and Practices (IBDAP) · doi:10.1109/ibdap65587.2025.11145826 · Cited by 5
“Guan et al. [32] enhanced this approach by integrating multiple LSTM layers, which led to improved performance compared to standalone LSTM models.”
Generalized Few-Shot Learning For Wearable Sensor-based Human Activity Recognition
Wijayasingha, L. N., & Stankovic, J. A. (2022) · IEEE International Conference on Pervasive Computing and Communications Workshops (PerCom Workshops) · doi:10.1109/percomworkshops53856.2022.9767306 · Cited by 7
“Lack of data for certain classes in HAR [15] can cause probles such as performance degradation, overfitting and reduced robustness [16].”
BackgroundFramework of Sequence Chunking for Human Activity Recognition Using Wearables
Zhang, W., Qin, L., Zhong, W., Guo, X., & Wang, G. (2019) · International Conference on Image, Video and Signal Processing (IVSP) · doi:10.1145/3317640.3317647 · Cited by 5
“Also, recurrent models [7,8,9] were adopted to model temporal dependencies.”
MethodPredictive Caching for AR/VR Experiences in a Household Scenario
Zehtabian, S., Razghandi, M., Bölöni, L. L., & Turgut, D. (2020) · 2012 International Conference on Computing, Networking and Communications (ICNC) · doi:10.1109/icnc47757.2020.9049692 · Cited by 5
“Majority voting is one of the basic prediction/classification methods in which multiple classifiers are used in order to predict the label based on the majority vote of the classifiers [17], [18].”
MethodWang, C., Wang, B., Liang, H., Zhang, J., Huang, W., & Zhang, W.-W. (2020) · IEEE Access · doi:10.1109/access.2020.2984456 · Cited by 5
Pellatt, L., & Roggen, D. (2020) · ACM International Joint Conference on Pervasive and Ubiquitous Computing (UbiComp) · doi:10.1145/3410530.3414365 · Cited by 5
Modeling an intelligent controller for predictive caching in AR/VR-enabled home scenarios
Zehtabian, S., Khodadadeh, S., Bölöni, L. L., & Turgut, D. (2021) · Pervasive and Mobile Computing · doi:10.1016/j.pmcj.2021.101334 · Cited by 5
Tian, A., Zhang, Y., Chen, H., Ma, C., & Zhou, S. (2021) · ACM International Joint Conference on Pervasive and Ubiquitous Computing (UbiComp) · doi:10.1145/3460418.3479383 · Cited by 5
“To tackle this problem, various deep learning networks have been proposed, such as CNNs [13], RNN [11], LSTMs[12] and GRUs.”
MethodCross-Domain Unseen Activity Recognition Using Transfer Learning
Khan, M. A. A. H., & Roy, N. (2022) · IEEE Annual International Computer Software and Applications Conference (COMPSAC) · doi:10.1109/compsac54236.2022.00117 · Cited by 6
Hiremath, S. K., & Ploetz, T. (2021) · 2021 International Symposium on Wearable Computers · doi:10.1145/3460421.3478825 · Cited by 4
“Training, validation and testing protocols are adopted from [11] for Opportunity (with 113 instead of 79 channels), PAMAP2, and Skoda.”
BackgroundMethodInfluentialSelf-Supervised Learning from Unlabeled IoT Data
Liu, D., & Abdelzaher, T. (2023) · Springer eBooks · doi:10.1007/978-3-031-40787-1_2 · Cited by 5
Gonçalves, H. R., Pinheiro, P., Pinheiro, C., Martins, L. M., Rodrigues, A. M., & Santos, C. P. (2025) · Computers in Biology and Medicine · doi:10.1016/j.compbiomed.2025.109961 · Cited by 4
Anzengruber-Tánase, B., Sopidis, G., Haslgrübler, M., & Ferscha, A. (2022) · International Conference on PErvasive Technologies Related to Assistive Environments (PETRA) · doi:10.1145/3529190.3529200 · Cited by 4
“This problem has been observed with real world data in the past and experiences with this issue are for example discussed in [8, 23, 28, 29].”
BackgroundZhao, T., Chen, H., Bai, Y., Zhao, Y., & Zhao, S. (2022) · International Journal of Environmental Research and Public Health · doi:10.3390/ijerph191811706 · Cited by 4
“[15] Developed modified training procedures for LSTM networks and combine sets of diverse LSTM learners into classifier collectives wearable sensors close/open dishwasher, close/open drawer, close/open door, close/open fridge, toggle switch, drink from cup, clean table”
BackgroundMethodParticle Swarm Optimization of Convolutional Neural Networks for Human Activity Prediction
Devarakonda, P. G., & Božić, B. (2022) · IntechOpen eBooks · doi:10.5772/intechopen.97259 · Cited by 8
Human Activity Recognition Using Machine Learning
Kaur, K., & Batth, J. S. (2024) · OPJU International Technology Conference on Smart Computing for Innovation and Advancement in Industry (OTCON) · doi:10.1109/otcon60325.2024.10688067 · Cited by 5
“To examine the recognition abilities for real-world unbalanced data, both DL network and DL networks have been given into model to test [17].”
MethodPredicting Daily Physical Activity Level for Older Adults Using Wearable Activity Trackers
Zheng, Y., Xie, J., Vo, T. V. T., Lee, B. C., & Ajisafe, T. (2019) · Lecture notes in computer science · doi:10.1007/978-3-030-22015-0_47 · Cited by 4
A New Motion Data Structuring for Human Activity Recognition Using Convolutional Neural Network
Alemayoh, T. T., Lee, J. H., & Okamoto, S. (2020) · IEEE/RAS-EMBS International Conference on Biomedical Robotics and Biomechatronics (BioRob) · doi:10.1109/biorob49111.2020.9224310 · Cited by 4
“Edel et al [11] and Guan et al [12], are among the works conducted using RNN.”
MethodMulti-class Multi-label Classification for Cooking Activity Recognition
Gashi, S., Lascio, E. D., & Silvia Santini, B. (2020) · Smart innovation, systems and technologies · doi:10.1007/978-981-15-8269-1_7 · Cited by 4
KATN: Key Activity Detection via Inexact Supervised Learning
You, X., Zhang, L., Yu, H., Yuan, M., & Li, X. (2021) · Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies · doi:10.1145/3494957 · Cited by 5
“Existing methods usually train a classification model for human activities relying on a collection of accurately labeled IMU data [48][4][29][43][17].”
Method(2019) · International Journal of Innovative Technology and Exploring Engineering · doi:10.35940/ijitee.f1111.0486s419 · Cited by 5
Cross-Domain Activity Recognition Using Shared Representation in Sensor Data
Hamad, R. A., Yang, L., Woo, W. L., & Wei, B. W. (2022) · IEEE Sensors Journal · doi:10.1109/jsen.2022.3178083 · Cited by 6
“Deep learning approaches in the form of ensemble learning are also used for activity recognition using wearable sensors [25].”
Multi-level Contrast Network for Wearables-based Joint Activity Segmentation and Recognition
Xia, S., Chu, L., Pei, L., Yu, W., & Qiu, R. C. (2022) · Globecom · doi:10.1109/globecom48099.2022.10001226 · Cited by 5
“The researchers utilized deep neural networks(DNNs) [11], convolutional neural networks(CNNs) [12], recurrent neural networks(RNNs) [13] or hybrid structures [14] to extract features from the related time series and then effectively recognized corresponding activities.”
BackgroundDitthapron, A., Lammert, A. C., & Agu, E. (2023) · IEEE Sensors Journal · doi:10.1109/jsen.2023.3290820 · Cited by 4
“…has focused primarily on improving HCR accuracy by addressing important context classification and data quality issues including missing labels [16], weakly labeled sensor data [17], [18], user/device-adaptation [19], [20], context class imbalance [21], noisy labels [1], [21] and DA [22], [23].”
BackgroundForecast of Broad Money Supply Based on Long Short-term Memory Neural Network (LSTM-NN)
Zhai, W., Wu, G., Xin, L., & Niu, L. (2021) · International Conference on Big Data and Information Analytics (BigDIA) · doi:10.1109/bigdia53151.2021.9619681 · Cited by 4
Wang, X., Wang, Y., Lu, C., Yu, H., He, H., & Li, Z. (2021) · International Conference on Advanced Mechatronic Systems · doi:10.1109/icamechs54019.2021.9661535 · Cited by 4
Ballroom Dance Movement Recognition Using a Smart WatchPreprint
Krishna, V. B. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2008.10122 · Cited by 4
“For human activity recognition, ensembles of deep LSTM networks were proposed in Guan & Pl¨otz (2017), but this approach is not suitable for real-time prediction because it is too slow.”
BackgroundPrivate, Fair and Secure Collaborative Learning Framework for Human Activity Recognition
Roy, D., Lekssays, A., Girdzijauskas, Š., Carminati, B., & Ferrari, E. (2023) · ACM International Joint Conference on Pervasive and Ubiquitous Computing (UbiComp) · doi:10.1145/3594739.3610675 · Cited by 3
“FL has been successfully applied in the ubiquitous Human Activity Recognition (HAR) scenario, where different users generate data from their wearable sensors [4, 8, 14].”
Ahmad, N., Leung, H.-F., & Farnia, F. (2025) · IEEE Internet of Things Journal · doi:10.1109/jiot.2025.3555799 · Cited by 4
“To evaluate the model under the hold-out approach, we follow studies [13], [20] in the distribution of data splitting for training, validation, and testing.”
InfluentialDeep Learning-Based Human Activity Recognition Using Binary Ambient Sensors
Zhao, Q., Ghasemi, A., Saif, A., & Bossard, L. (2026) · Electronics · doi:10.3390/electronics15020428 · Cited by 3
Assessment of IMU Configurations for Human Activity Recognition Using Ensemble Learning
Murphy, S. J., & Vitali, R. V. (2024) · IEEE Access · doi:10.1109/access.2024.3440215 · Cited by 3
“Ensemble classifiers has been shown to produce stronger classifiers that are more generalizable [22], [30].”
BackgroundMethodSmart Assisted Living: Toward An Open Smart-Home Infrastructure
Gao, Y., Long, Y., Guan, Y., Basu, A. P., Baggaley, J., & Plötz, T. (2019) · Cited by 3
Self-citation
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.