Deep Convolutional and LSTM Recurrent Neural Networks for Multimodal Wearable Activity Recognition
Open accessThe DOI matches its Crossref record; 48 of the 49 references it cites were found in a registry or an index; 8 of the 23 sources its claims cite were checked for support.
Abstract
Introduces DeepConvLSTM, a deep learning architecture combining convolutional and LSTM recurrent units for multimodal wearable activity recognition, outperforming non-recurrent deep networks and handcrafted feature baselines by 4% to 9%.
Overview
CoChat Summary
TL;DR
The paper proposes DeepConvLSTM, a deep framework combining four convolutional layers with two LSTM recurrent layers for wearable activity recognition from raw multimodal sensor data. It reports improved F1 scores over a baseline CNN and prior results on the OPPORTUNITY and Skoda datasets, including 0.915 F1 for OPPORTUNITY gestures with the Null class and 0.958 F1 on Skoda.
Key findings
- On OPPORTUNITY, DeepConvLSTM achieved F1 scores of 0.930 for modes of locomotion without Null, 0.895 for modes with Null, 0.866 for gestures without Null, and 0.915 for gestures with Null; the baseline CNN scored 0.912, 0.878, 0.783, and 0.883, respectively. The paper reports a 6% average improvement over the best OPPORTUNITY challenge submissions and more than 9% improvement on gesture recognition without Null.
- On Skoda, DeepConvLSTM reached 0.958 F1, compared with 0.884 for the baseline CNN, 0.893 for CNN [43], and 0.861 for CNN [23], improving the best reported result by 6%.
- Multimodal fusion improved gesture recognition on OPPORTUNITY without Null: 0.689 F1 with accelerometers only, 0.611 with gyroscopes only, 0.745 with accelerometers + gyroscopes, 0.839 with accelerometers + gyroscopes + magnetic sensors, and 0.864 with the full OPPORTUNITY sensor set.
- DeepConvLSTM had 999,122 parameters versus 7,445,458 for the baseline CNN, a 600% increase for the baseline; training took 340.3 min versus 282.2 min, and classification took 6.68 s versus 5.43 s, with DeepConvLSTM classifying almost 15 min of data per second and achieving 900× real-time on the GPU used.
Methods
- DeepConvLSTM uses four convolutional layers with 64 feature maps each, followed by two dense LSTM layers with 128 cells and a softmax output; the baseline CNN shares the convolutional layers but uses two non-recurrent dense layers with 128 units. Models were trained supervised with cross-entropy loss, mini-batch gradient descent with RMSProp, learning rate 10e-3, decay factor 0.9, mini-batches of 100 segments, and dropout p=0.5 on dense-layer inputs.
- Inputs were sliding windows of 500 ms with 250 ms step size, using raw sensor data with minimal preprocessing: linear interpolation for missing values and per-channel normalization to [0,1]. Evaluation used weighted F1 to address class imbalance.
- Experiments used the OPPORTUNITY dataset (on-body sensors, 113 channels, 30 Hz; Task A: 5-class modes of locomotion/postures; Task B: 18-class sporadic gestures) and the Skoda dataset (one subject, 10 right-arm 3D accelerometers, decimated to 30 Hz, 10 manipulative gestures, about 3 h and about 70 repetitions per gesture). Analyses included comparison to baseline CNN and published results, confusion matrices, sensor-modality fusion, and hyperparameter evaluations of sequence length and number of convolutional layers.
Limitations
- The paper states that DeepConvLSTM has not yet been implemented on a wearable device; the GPU used clearly outperforms current high-end wearable computational power, and training is best envisioned server-side. It also states that pooling was omitted because the sliding-window setup limits downsampling, though pooling could be useful without that requirement, and that training time increases with the number of layers, creating a performance–computational budget trade-off.
- Evident from the text: Skoda evaluation uses one subject and only right-arm sensors, and OPPORTUNITY testing is limited to ADL4 and ADL5 for Subjects 2 and 3, which constrains generalizability. The text does not report cross-validation or statistical significance testing.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Figures



![Figure 4 Placement of on-body sensors used in the OPPORTUNITYdataset (left: inertial measurements units; right: 3-axis accelerometers) [7].](https://figures.cochat.ai/paper-figures/27c0e4b39d104fe2b5315587d736fd2d/deep-convolutional-and-lstm-recurrent-neural-networks-for-multimodal-wearable-figure-4.jpg)




Verification
Claims the paper makes
5 Valid · 5 Not valid · 13 Not checked · 99 Own results and interpretations
Every source the paper cites for this claim was checked and supports it.
In the paper
“The choice of the number of recurrent layers is made following the results presented in [33], where the authors showed that a depth of at least two recurrent layers is beneficial when processing sequential data.”
§ Deep Convolutional and LSTM Recurrent Neural Networks for Multimodal Wearable Activity Recognition > 3. Architecture > 3.1. DeepConvLSTM
What the sources say
- Supportedcites [33] Karpathy A. Johnson J. Li F.F. Visualizing and understanding recurrent networks Cornell Univ. Lab. 2015 arXiv:1506.02078 source
Section 4.1 compares recurrent networks with one, two, and three layers on two sequence datasets and states that a depth of at least two is consistently beneficial. This directly supports the attributed finding, within the paper’s tested settings. (LLM, conf=0.94)
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.
- Supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“This seems to indicate that although deep learning techniques are often employed with large amounts of data, (e.g., millions of frames in computer vision [22]), they may actually be applicable to problem domains where acquiring annotated data is very costly, such as in supervised activity recognition.”
§ Deep Convolutional and LSTM Recurrent Neural Networks for Multimodal Wearable Activity Recognition > 5. Results and Discussion > 5.4. Discussion
What the sources say
- Supportedcites [22] Pigou L. Oord A.V.D. Dieleman S. van Herreweghe M. Dambre J. Beyond Temporal Pooling: Recurrence and Temporal Convolutions for Gesture Recognition in Video arXiv Preprint 2015 arXiv:1506.01911 source
The citation is attached to the example of large-scale computer-vision data, not to the citing authors’ conclusion about costly annotation. Section 4.1 (page unavailable) reports 1,720,800 labeled video frames in the gesture-recognition dataset, supporting the cited example of data on the order of millions of frames. (LLM, conf=0.86)
hedgedstates a possible 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.
- Supported
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
“The baseline CNN also offers better results than the OPPORTUNITY submissions in the recognition of models of locomotion.”
§ Deep Convolutional and LSTM Recurrent Neural Networks for Multimodal Wearable Activity Recognition > 5. Results and Discussion > 5.1. Performance Comparison
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether 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
“From the results, we can also see how the model is able to learn from signals obtained from accelerometers, gyroscopes and magnetometers, fusing them without requiring any specific preprocessing.”
§ Deep Convolutional and LSTM Recurrent Neural Networks for Multimodal Wearable Activity Recognition > 6. Conclusions
plainstates 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.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“(iv) it is applicable even if the gestures are longer than the observation window.”
§ Deep Convolutional and LSTM Recurrent Neural Networks for Multimodal Wearable Activity Recognition > 5. Results and Discussion > 5.4. Discussion
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: unclear — It is unclear why the paper says it holds.
84 more claims
Create a free account to see all 122 claims and to verify the 2 references still open.
- It may state two separate claims as one. 14
- Probably not a substantial claim. 23
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 doesn’t support the claim as written.
In the citing paper
“However, these complex network framework suffered from low ef ciency and can hardly meet real-time requirements in practice applications.”
From InnoHAR: A Deep Neural Network for Complex Human Activity Recognition (2019) · cites this paper as [19]
Checked by a reasoning model — may be wrong.
Experimental — accuracy not yet measuredThe cited source doesn’t support the claim as written.
In the citing paper
“However, due to the lack of filtration techniques applied to multimodal data, the system’s performance was low.”
From A Multimodal IoT-Based Locomotion Classification System Using Features Engineering and Recursive Neural Network (2023) · cites this paper as [15]
Checked by a reasoning model — may be wrong.
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“and yield better performances than pure CNNs or LSTM networks [12]”
From Comparison of Feature Learning Methods for Human Activity Recognition Using Wearable Sensors (2018) · cites this paper as [12]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“The proposed method outperforms other methods, such as those based on deep believe networks (DBNs) [10], SVM [10], and CNNs [13].”
From Deep Recurrent Neural Networks for Human Activity Recognition (2017) · cites this paper as [13]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“The popular DeepConvLSTM model [7] can only afford four, moderately-sized convolution layers to learn effective data representations, and two LSTM layers for sequence modeling.”
From Complex Deep Neural Networks from Large Scale Virtual IMU Data for Effective Human Activity Recognition Using Wearables (2021) · cites this paper as [7]
Experimental — accuracy not yet measured
20 more claims about this paper
Create a free account to see every claim other papers make about this one.
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 major: it would touch a result — a value a result rests on, reported and recomputed p-values on opposite sides of the significance level, two passages that conflict with no sign which is right, or a cited work retracted before this paper was published.
In the paper
§ Deep Convolutional and LSTM Recurrent Neural Networks for Multimodal Wearable Activity Recognition > 3. Architecture
“The length of the kernels is the same for every convolutional layer, being defined as Pl=5,∀l=2,⋯,5.”
§ Deep Convolutional and LSTM Recurrent Neural Networks for Multimodal Wearable Activity Recognition > 4. Experimental Setup > 4.2. Performance Measure
“Following the OPPORTUNITY challenge experimental setup, the length of the window is 500 ms, with a step size of 250 ms.”
Rated major: it would touch a result — a value a result rests on, reported and recomputed p-values on opposite sides of the significance level, two passages that conflict with no sign which is right, or a cited work retracted before this paper was published.
In the paper
“| Total | 996,800 +(128×nc)+nc | 996,800 +(128×nc)+nc | 7,443,136+(128×nc)+nc | 7,443,136+(128×nc)+nc |”
§ Deep Convolutional and LSTM Recurrent Neural Networks for Multimodal Wearable Activity Recognition > 3. Architecture > 3.3. Model Implementation and Training
Reported 996800 · recomputed 1037888
Rated major: it would touch a result — a value a result rests on, reported and recomputed p-values on opposite sides of the significance level, two passages that conflict with no sign which is right, or a cited work retracted before this paper was published.
In the paper
“| 6 | Wai,Waf,Wac,Wao: 7232×128 | 942,592 | W: 57,856×128 | 7,405,696 |”
§ Deep Convolutional and LSTM Recurrent Neural Networks for Multimodal Wearable Activity Recognition > 3. Architecture > 3.3. Model Implementation and Training
Reported 942592 · recomputed 3702784
Rated major: it would touch a result — a value a result rests on, reported and recomputed p-values on opposite sides of the significance level, two passages that conflict with no sign which is right, or a cited work retracted before this paper was published.
In the paper
“| 7 | Wai,Waf,Wac,Wao: 128×128 | 33,280 | W: 128×128 | 16,512 |”
§ Deep Convolutional and LSTM Recurrent Neural Networks for Multimodal Wearable Activity Recognition > 3. Architecture > 3.3. Model Implementation and Training
Reported 33280 · recomputed 131072
Citations
Cites
ImageNet classification with deep convolutional neural networks
Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2017) · Communications of the ACM · doi:10.1145/3065386 · Cited by 112,155
Going deeper with convolutions
Szegedy, C., Liu, W., Jia, Y., Sermanet, P., Reed, S., Anguelov, D., Erhan, D., Vanhoucke, V., & Rabinovich, A. (2015) · 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2015.7298594 · Cited by 48,562 · cited 1×
“Other publications showed much deeper networks, such as “GoogLeNet”, which is a 27-layer deep neural network applied to image classification [45].”
BackgroundReducing the dimensionality of data with neural networks
Hinton, G. E., & Salakhutdinov, R. R. (2006) · Science · doi:10.1126/science.1127647 · Cited by 20,928
Speech recognition with deep recurrent neural networks
Graves, A., Mohamed, A.-R., & Hinton, G. (2013) · 2013 IEEE International Conference on Acoustics, Speech and Signal Processing · doi:10.1109/icassp.2013.6638947 · Cited by 8,893 · cited 1×
“Networks using LSTM cells have offered better performance than standard recurrent units in speech recognition, where they gave state-of-the-art results in phoneme recognition [20].”
BackgroundConvolutional networks for images, speech, and time series
LeCun, Y., & Bengio, Y. (1998) · HAL (Le Centre pour la Communication Scientifique Directe) · doi:10.5555/303568.303704 · Cited by 4,231
The class imbalance problem: A systematic study1
Japkowicz, N., & Stephen, S. (2002) · Intelligent Data Analysis · doi:10.3233/ida-2002-6504 · Cited by 3,295
DeepPose: Human Pose Estimation via Deep Neural Networks
Toshev, A., & Szegedy, C. (2014) · 2014 IEEE Conference on Computer Vision and Pattern Recognition · doi:10.1109/cvpr.2014.214 · Cited by 3,185 · cited 1×
“Deep CNNs have had a major impact on fields, like content recommendation [24], speech recognition [25] and in computer vision [26–28], where they have become the de facto standard approach.”
BackgroundOverFeat: Integrated Recognition, Localization and Detection using Convolutional Networks
Sermanet, P., Eigen, D., Zhang, X., Mathieu, M., Fergus, R., & LeCun, Y. (2013) · arXiv · doi:10.48550/arxiv.1312.6229 · Cited by 2,680 · cited 1×
“Deep CNNs have had a major impact on fields, like content recommendation [24], speech recognition [25] and in computer vision [26–28], where they have become the de facto standard approach.”
BackgroundLee, H., Grosse, R., Ranganath, R., & Ng, A. Y. (2009) · Proceedings of the 26th Annual International Conference on Machine Learning · doi:10.1145/1553374.1553453 · Cited by 2,412 · cited 1×
“Deep learning techniques have outperformed many conventional methods in computer vision [13] and audio classification [14].”
MethodA review of wearable sensors and systems with application in rehabilitation
Patel, S., Park, H., Bonato, P., Chan, L., & Rodgers, M. (2012) · Journal of NeuroEngineering and Rehabilitation · doi:10.1186/1743-0003-9-21 · Cited by 2,396 · cited 1×
“, from simple hand gestures to complex activities, such as “cooking a meal”) and the context in which they occur from sensor data is at the core of smart assistive technologies, such as in smart homes [1], in rehabilitation [2], in health support [3,4], in skill assessment [5] or in industrial settings [6].”
Background
Research works printed but not linked: 1
- Gers F.A. Schraudolph N.N. Schmidhuber J. Learning precise timing with LSTM recurrent networks J. Mach. Learn. Res. 2003 3 115 143
Cited by
Items 2,601–2,700 of 2,807
Visualizing and Analyzing Human Activity Patterns through Graph-Based Methods
Rahin, S. A., Hui, B., & Li, W. (2024) · International Conference on Machine Learning and Machine Intelligence (MLMI) · doi:10.1145/3696271.3696310
“To address this, researchers have explored various sequence modeling techniques such as Hidden Markov Models (HMM) [11], and Recurrent Neural Networks (RNN) [12].”
Zhao, J., & Obonyo, E. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2008.01679
“Current ML-based recognition models were developed from the full posture datasets of workers’ in related studies [4-10, 12].”
BackgroundMethodInfluentialYu, Y., Dong, W., Dong, H., Wang, W., Gao, Y., Wu, D., & Lin, W. (2026) · Sensors · doi:10.3390/s26123706
An automated human behavior analysis using AI based predictive model in health care
Dutta, P., Paul, S., Dey, S., Das, R., & Jana, G. G. (2024) · Elsevier eBooks · doi:10.1016/b978-0-443-15452-2.00003-0
Chai, Z., Zhou, Y., & Yu, H. (2026) · International Conference on Automation and Computing (ICAC) · doi:10.1109/icac70325.2026.11712318
Alanazi, M., Alsharif, B., Altaher, A. S., Altaher, A., & Ilyas, M. (2026) · Franklin Open · doi:10.1016/j.fraope.2026.100728
Development of the Application to Promote Daily Physical Activities in Adolescents
Oļehnovičs, E., Laizāns, A., Kļaviņa, A., et al. (2022) · Human, Technologies and Quality of Education · doi:10.22364/htqe.2022.80
Classification de données massives de télédétection
Audebert, N. (2018) · theses.fr (ABES) · doi:10.70675/e7e6756dzea11z49c1z99dfz2a794a46b421 · Cited by 4
Migliato, A. L. T. (2021) · Digital Library of Theses and Dissertations (Universidade de São Paulo) · doi:10.11606/d.55.2021.tde-21012022-175531 · Cited by 2
Kostolani, D., Wolling, F., Schlund, S., & Michahelles, F. (2026) · Frontiers in Computer Science · doi:10.3389/fcomp.2026.1873283
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
“Following the splitting protocol outlined in [39], we designated the data from the third ADL session of subject 2 and subject 3 as the validation set.”
Description of Structural Biases and Associated Data in Sensor-Rich EnvironmentsPreprint
Hamidi, M., & Osmani, A. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2104.04885
“[136, 137, 138, 139], these kinds of networks are being adopted, primarily, for their ability to efficiently aggregate heterogeneous data from different sources.”
BackgroundZuo, J., Peng, X., Sun, X., & Wang, Q. (2026) · IEEE Sensors Journal · doi:10.1109/jsen.2026.3712390
“However, most existing IMU-based motion understanding methods still formulate the problem as fixed-window clip-level classification, where inertial sequences are first divided into fixed-length clips and each clip is assigned an activity or motion label [2], [4], [5], [6], [7], [8], [9], [10].”
InfluentialGonzalez, D. R., & Qu, Y. (2023) · Congress in Computer Science, Computer Engineering, and Applied Computing (CSCE) · doi:10.1109/csce60160.2023.00084 · Cited by 1
“In recent years, with the rapid development of deep learning technology, the classification performance of human activity recognition based on deep learning networks has increased substantially [6].”
BackgroundA Conventional Method for Deep Learning-Based Human Activity Recognition Analysis
Gayathri, B., Kumar, V. N., Shanthi, B., Rafeeq, M. D., & Ibrahim, M. M. (2026) · Smart innovation, systems and technologies · doi:10.1007/978-3-032-12830-0_49
A Multibranch LSTM-CNN Model for Human Activity Recognition
Kabir, S., Ahmed, M. U., Hossen, M. M., & Chowdhury, M. E. H. (2024) · Studies in computational intelligence · doi:10.1007/978-3-031-71821-2_6
Towards Intelligent Security: Embedded AI and Multi-Sensor Data Fusion for Threat Detection
Lazrak, Z., Bouayad, A., & Boumaiz, M. (2025) · International Conference on Embedded Systems and Artificial Intelligence (ESAI) · doi:10.1109/esai67033.2025.11438584
“All the parameters and layer configurations are listed in TABLE 3 [19][20].”
Esfahani, A. H., Maye, O., Frohberg, M., Speh, M., Jöbges, M., & Langendörfer, P. (2024) · Procedia Computer Science · doi:10.1016/j.procs.2024.06.160
Assessment of ML Classifiers in Complex Human Activity Recognition Using Wearable Sensors Data
Hamid, S., Awan, M. A., & Saeed, M. (2024) · International Journal of Innovations in Science and Technology · doi:10.33411/ijist/1088
Lightweight Deep Learning for Sensor-Based HAR: Benchmarking Optimization Strategies
Agac, S., & İncel, Ö. D. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-13312-0_5
Time‐Frequency Transformer: A Network for Sensor‐Based Human Activity Recognition
Xiao, S., Deng, J., Fang, C., Tang, C., & Huang, Z. (2026) · Computational Intelligence · doi:10.1111/coin.70225 · Cited by 1
Anomaly detection on MNIST stroke simulation dataset
Vishwakarma, O., Agarwal, S., Pandian, K. K. S., et al. (2022) · International Conference on Electrical Engineering, Computer Science and Informatics (EECSI) · doi:10.23919/eecsi56542.2022.9946546
“Also, the efficacy of LSTM in modelling motion data for activity recognition has been already demonstrated in the past [28].”
BackgroundLee, J., Kang, J., & Sohn, M. M. (2023) · Communications in computer and information science · doi:10.1007/978-981-99-4430-9_8
Almeida, L., Santos, V. M. F., & Ferreira, J. P. (2019) · Iberian Conference on Information Systems and Technologies (CISTI) · doi:10.23919/cisti.2019.8760604
Deep learning for time series classification
Fawaz, H. I. (2020) · theses.fr (ABES) · doi:10.70675/4be65ba1z3327z4573zb045z15c1d1435785 · Cited by 11
Mekruksavanich, S., Phaphan, W., & Jitpattanakul, A. (2026) · International Conference on Big Data Analytics and Practices (IBDAP) · doi:10.1109/ibdap69915.2026.11678004
Wang, H., She, T., Wu, Y., Jiang, L., Wang, X., & Yuan, Z. (2026) · ACS Applied Electronic Materials · doi:10.1021/acsaelm.6c00932
Xie, W., Huang, Q., Li, C., Wang, Y., & Liu, Y. (2023) · ACM International Conference on Multimedia in Asia (MMAsia) · doi:10.1145/3595916.3626425
“Ordóñez et al. [19] achieved significant results in capturing spatio-temporal features from sensor signals by combining CNN and LSTM. Zeng et al. [20] proposed sensor modality attention and temporal attention on top of a loop layer and added continuity constraints to sensor signals.”
BackgroundInfluentialReview of deep learning methods for imaging photoplethysmography data processing
Rubin, I. M., & Volynsky, M. A. (2026) · Scientific and technical journal of information technologies mechanics and optics · doi:10.17586/2226-1494-2026-26-2-236-249 · Cited by 1
Leveraging multi-dimensional context & data augmentation for NER
Luiggi, T. (2025) · theses.fr (ABES) · doi:10.70675/75128aa5z75b0z4644z96b1z04406ee442f3
Activity recognition using smartphones and wearable devices: Traditional approaches, New solutions
Iskanderov, J., & Guvensan, M. A. (2019) · Pamukkale University Journal of Engineering Sciences · doi:10.5505/pajes.2019.84758
Abbasi, M. H., Mehrabadi, F. A., & Bitaraf, M. (2026) · Structural Control and Health Monitoring · doi:10.1155/stc/3434617
Stephen, O., Madanian, S., & Nguyen, M. (2022) · International Conference on e-Business · doi:10.5220/0011296400003280
“Among these studies, a study involved space and time characteristics extraction (Ordóñez & Roggen, 2016) combing four layers of CNN and two layers of LSTM achieved a superior result compared to CNN only.”
BackgroundResultAdaptive Multi-state Pipe Framework Based on Set Pair Analysis
Shi, L., Xu, H., Zhang, B., Sun, X., Li, J., & Fan, S. (2021) · International Journal of Machine Learning and Computing · doi:10.18178/ijmlc.2021.11.2.1029
StarHAR: A Lightweight and Low-Latency Framework for Sensor-Based Human Activity Recognition
Song, M., & Feng, S. (2025) · Lecture notes in computer science · doi:10.1007/978-981-96-9875-2_29
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
Shuhui, G., Cao, M., & Wan, J. (2026) · International Journal of Data Science and Analytics · doi:10.1007/s41060-026-01212-8
A Computationally Light Semi-Supervised Learning Framework for Constrained Embedded Platforms
Roy, A., Bhuyan, A. K., & Biswas, S. K. (2025) · IEEE Internet of Things Journal · doi:10.1109/jiot.2025.3646121
“Similarly, the hierarchical architecture proposed in [24] introduces multitiered learning models that demand considerable memory and compute power, which is often incompatible with microcontroller-level constraints.”
Enhancing Health Monitoring using Efficient Hyperparameter Optimization
Singhal, R. (2022) · Journal of Artificial Intelligence and Capsule Networks · doi:10.36548/jaicn.2022.4.004
Haresamudram, H., Essa, I. A., & Plötz, T. (2024) · International Conference on Activity and Behavior Computing (ABC) · doi:10.1109/abc61795.2024.10651688 · Cited by 1
“Going beyond hand-crafting of features [18], [19], recent approaches learn useful representations from data itself, through supervised and unsupervised/self-supervised learning [1], [20].”
BackgroundA Categorical Kalman Filter for Human Activity Recognition
de França, D. S., Neto, A. R. P., Cruz, L. A., & Macedo, J. A. F. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-15987-8_1
Zeng, J., Xu, H., Gao, F., Sun, F., Wang, X., Li, X., Li, Y., Li, X., & Ai, W. (2025) · SSRN Electronic Journal · doi:10.2139/ssrn.5152077
Dominance of Deep LSTM in Smartphone Sensor based Human Activity Classification
Faiyaz, A., Awlad, M. S. I., Masood, N. A., et al. (2020) · IEEE Asia-Pacific Conference on Computer Science and Data Engineering (CSDE) · doi:10.1109/csde50874.2020.9411603
“Another study [13] indicated the approach of LSTM (Long short-term memory recurrent) and RNN on memory limitation problem of deep convolution learning.”
BackgroundUnified time series classification framework for explainable artificial intelligence
Bragança, H. L. D. S., & Souto, E. (2026) · Scientific Reports · doi:10.1038/s41598-026-49467-2 · Cited by 1
“To evaluate the effectiveness of our approach, we use three baseline models: DeepConvLSTM 24 , Fully Convolutional Networks (FCN) 25,26 , and XGBoost.”
Xue, S., & Fang, Y. (2022) · IEEE Conference on Telecommunications, Optics and Computer Science (TOCS) · doi:10.1109/tocs56154.2022.10016033
“In related studies, some researchers divided the transportation modes into bus, car, motorcycle, subway, or train [3].”
BackgroundMethodDeep Learning Transformer Model for Human Activity Recognition
Iftode, I.-A., & Foşalău, C. (2024) · Buletinul Institutului Politehnic din Iaşi. Secţia Electrotehnică. Energetică. Electronică · doi:10.2478/bipie-2024-0011
“…by Ordóñez and Roggen, integrating an LSTM layer atop CNN-derived features elevated the F1 score from 91.8% to 94.1% on the Opportunity dataset, underscoring the efficacy of hybrid architectures in capturing both spatial and temporal characteristics simultaneously (Ordóñez and Roggen, 2016).”
InfluentialPredicting short‐term mobile Internet traffic from Internet activity using recurrent neural networks
Santos, G. L., Rosati, P., Lynn, T., et al. (2021) · International Journal of Network Management · doi:10.1002/nem.2191 · Cited by 15
Koo, H. (2026) · Applied Sciences · doi:10.3390/app16157476
Physics-informed hierarchical transformer for wearable sensor-based gait fatigue assessment
Ding, K., Li, Z., Zou, X., Jia, J., Li, C., & Jiang, Z. (2026) · Frontiers in Public Health · doi:10.3389/fpubh.2026.1794241
A Public Repository to Improve Replicability and Collaboration in Deep Learning for HAR
Pellatt, L., Bock, M., Roggen, D., & Laerhoven, K. V. (2022) · IEEE International Conference on Pervasive Computing and Communications Workshops (PerCom Workshops) · doi:10.1109/percomworkshops53856.2022.9767436
“To investigate whether grokking can be also exhibited within HAR we trained the DeepConvLSTM model as proposed by Ordonez and Roggen [1] for 300 epochs.”
BackgroundMethodResultInfluentialSelf-citationBuvan, S., Nithesh, R., & Nithya, A. (2025) · 2025 International Conference on Data Science, Agents & Artificial Intelligence (ICDSAAI) · doi:10.1109/icdsaai65575.2025.11011758
Madaeni, F., Chokmani, K., Lhissou, R., et al. (2021) · The Cryosphere · doi:10.5194/tc-2021-194-rc1
Li, Q., Qu, Z., Yan, M. L., & Zhang, X. (2026) · Sensors · doi:10.3390/s26144390
“DeepConvLSTM, convolutional/recurrent comparison studies, and ensemble LSTM models demonstrate that raw or lightly processed wearable sequences can support end-to-end feature learning [2–4].”
Exploring Cross Modality Feature Fusion for Activity Recognition at OpenPack Challenge 2022
Inoshita, T., Namba, Y., Nakatani, Y., et al. (2023) · IEEE International Conference on Pervasive Computing and Communications Workshops and other Affiliated Events (PerCom Workshops) · doi:10.1109/percomworkshops56833.2023.10150371
“For example, 3DCNN [2] combines spatial and temporal information video analysis, ST-GCN [3] recognizes actions based on the motion of key points on the human body, and DeepConvLSTM [4] is often used for time series analysis of wearable sensors.”
MethodKresevic, S., Miladinović, A., ACCARDO, A., et al. (2026) · World Congress on Medical Physics and Biomedical Engineering, September 7 - 12, 2009, Munich, Germany · doi:10.1007/978-3-032-37736-4_14
FLAP: Federated Learning with Attack and Privacy AwarenessPreprint
Xue, W., Ahmed, N., Masood, R., Hu, W. S., Seneviratne, A. P., & Jha, S. K. (2022) · doi:10.36227/techrxiv.19766023
Modeling Affective Distress from Video: A Modality-Aware Multi-Branch Learning Approach
Jha, O. G., Sahu, N. K., & Lone, H. R. (2026) · INTERNATIONAL CONFERENCE ON MULTIMODAL INTERACTION · doi:10.1145/3776574.3831196
Mask-Aware Dynamic Gating for Robust Multi-Sensor Activity Recognition
Ouyang, W., Liu, S., Meng, T., Zhou, X., & Yang, Y. (2026) · International Conference on Big Data and Information Analytics (BigDIA) · doi:10.1109/bigdia71496.2026.11684722
“DeepConvLSTM is a representative architecture that combines convolutional feature extraction with recurrent sequence modeling for multimodal wearable activity recognition [8].”
Efficient Sparse Processing in Smart Home Applications
Chandrasekaran, R., Guo, Y., Thomas, A. H., Menarini, M., Ostertag, M. H., Kim, Y., & Rosing, T. (2019) · Proceedings of the 1st Workshop on Machine Learning on Edge in Sensor Systems · doi:10.1145/3362743.3362963
“While we again stress that our goal is not to compare the accuracy of our approach with prior work, we remark that our accuracy results are comparable to 73% - 94% achieved by state-of-the-art as in [8].”
ResultPersonalization Models for Human Activity Recognition with Distribution Matching-Based Metrics
Nguyen, H. T., Kwon, H., Haresamudram, H., Peterson, A. F., & Plötz, T. (2021) · Communications in computer and information science · doi:10.1007/978-981-16-0575-8_4
Poster: Attention-based Spatio-Temporal Model for HAR Using Multivariate Time Series
Xi, R., Li, M., Liu, D., et al. (2019) · International Conference on Embedded Wireless Systems and Networks · doi:10.5555/3324320.3324360
yang, Z., Li, L., Cai, Y., Sun, Z., Tao, L., Bao, C., Meng, Q., Li, X., Xue, Y., Xiong, K., et al. (2026) · Medical Engineering & Physics · doi:10.1088/1873-4030/aea226
“DeepConvLSTM-style architectures combine convolutional feature extraction with recurrent temporal modeling and have become influential baselines for wearable sensor activity recognition (Hammerla et al 2016, Ordóñez and Roggen 2016).”
Phát 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
A Real-World Dataset for detecting Handwashing in daily Life using Wrist Motion Data from Wearables
Burchard, R., Kirsten, K., Miché, M., Scholl, P., Arnrich, B., Laerhoven, K. V., Lieb, R., & Wahl, K. (2026) · Scientific Data · doi:10.1038/s41597-026-06698-2 · Cited by 3
“Cao et al . employed a CNN-LSTM network similar to DeepConvLSTM 24 to distinguish gestures related to handwashing and recorded a small dataset for their study, which was not available online 25 .”
InfluentialHuan, R., Zhou, Y., Dong, G., Chen, P., Liang, R., & Chen, C. (2026) · Pattern Analysis and Applications · doi:10.1007/s10044-026-01750-8
“Ordóñez et al. [ 17 ] proposed DeepConvLSTM, where convolutional operations extract local sensor patterns and LSTM components capture temporal dependencies in wearable sequences.”
Alonazi, M., Abro, I. A., Abdelhaq, M., Alsaqour, R. A., Jalal, A., & Liu, H. (2026) · Bioengineering · doi:10.3390/bioengineering13070718 · Cited by 3
“Thus, for example, Ordóñez & Roggen [13] employed a deep convolutional and LSTM recurrent architecture (DeepConvLSTM) for human activity recognition.”
Learning and mining from personal digital archives
Li, N. (2020) · Dublin City University Open Access Institutional Repository (Dublin City University)
Beyond the Motion-Lab: Low-Cost Fatigue Monitoring in Cycling via Deep Learning
Marquez, L. E. N., Garzón, G. O., Portilla, C. C.-D. L., et al. (2026) · IEEE Colombian Conference on Applications of Computational Intelligence (ColCACI) · doi:10.1109/colcaci71483.2026.11691351
“Both LSTM variants [14], [15] improve on this base-line, but the comparison between them is more nuanced than it first appears: although the attention-based model [16] achieved the highest single-seed accuracy (78.7%), when averaged across five random seeds the plain LSTM (65.4”
InfluentialPrincipled change point detection via representation learning.Preprint
Romanenkova, E., Zaytsev, A., Zainulin, R., & Morozov, M. (2021) · arXiv (Cornell University)
Cross-Sensor Masked Self-Supervised Transformers for Domain-Adaptive Human Activity Recognition
Karim, M. M., Kabir, S., & Fattah, S. A. (2026) · 2026 IEEE Region 10 Symposium (TENSYMP) · doi:10.1109/tensymp68909.2026.11704026
“Convolutional, recurrent, and hybrid CNN-LSTM architectures have demonstrated strong performance when training and testing data are collected under similar conditions [1], [3].”
A Fused Modality Human Action Recognition System Based on Motion Saliency in RGBD Videos
Gutev, A., & Debono, C. J. (2025) · Cognitive Models and Artificial Intelligence Conference (AICCONF) · doi:10.1109/aicconf64766.2025.11064223
“HARclassifiers trained on data from multiple modalities, such as skeleton joint information, depth and RGB color information modalities, have demonstrated a higher classification accuracy than those relying on a single modality [4], [5].”
Sawsen, G., Ouarda, W., Boubaker, H., Khelifa, M. M. B., & Alimi, A. M. (2021) · doi:10.36227/techrxiv.13991198.v1
Kalaivaani, P. C. D., S, H. L., Janarthikan, V., & R, J. S. (2026) · International Conference on Pervasive Computing and Social Networking (ICPCSN) · doi:10.1109/icpcsn68523.2026.11543823
“Additionally to vision-based HAR, Ordonez and Roggen [12] suggested a deep CNN-LSTM architecture on multi-modal wearable sensor data, which achieves better recognition through shared spatial and temporal features of data representation, which also improved recognition accuracy.”
Lovanshi, M., Das, S., & Tiwari, V. (2026) · IEEE Sensors Journal · doi:10.1109/jsen.2026.3713819 · Cited by 1
Human Activity Recognition Model Based on the Fusion of Capsule-Transformer
王, 星. (2025) · Modeling and Simulation · doi:10.12677/mos.2025.145466
ACTIVITY RECOGNITION WITH A HYBRID CNN–NCP AND SELF-ATTENTION-BASED MODEL
Olgun, N., Çalışan, M., & Balım, C. (2025) · Middle East Journal of Science · doi:10.51477/mejs.1667055
Wu, Q., Chu, S., Li, K., & Wang, L. (2026) · Applied Sciences · doi:10.3390/app16126037
Liu, C., Liu, X., Mi, S., Liu, Y., Tian, Y., Zhang, Y., Zhang, X., & Pang, L. (2026) · Frontiers in Physiology · doi:10.3389/fphys.2026.1911268
Nikhil, S., Samanta, S., & Kumaran, R. S. (2026) · International Conference on Wireless Communications, Signal Processing and Networking (WiSPNET) · doi:10.1109/wispnet69615.2026.11489421
“Hybrid CNN-LSTM architectures overcome this [16], [17]: Ali et al. [11] applied them to 5G NOMA detection (~85% test accuracy) and Devi et al. [14] achieved 91.2% on 5G signals.”
InfluentialTop-Confidence Gapped Cross-Entropy for Compact Human Activity Recognition
Alzhrani, K. (2026) · Applied Sciences · doi:10.3390/app16073394
Runtime Heterogeneous Sensor Selection with Data Reuse in Multi-device Environments
Li, C. L., Zhang, Y., Khyzer, H., Yan, Y., & Zhou, X. (2026) · Communications in computer and information science · doi:10.1007/978-981-95-2581-2_20
Research on SQL Injection Defense Technology Based on Deep Learning
Shi, W., & Liu, X. (2022) · Lecture notes in computer science · doi:10.1007/978-3-031-06788-4_45
Training neural networks with synthetic electrocardiogramsPreprint
Kaisti, M., Laitala, J., & Airola, A. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2111.06175 · Cited by 2
“Furthermore, the properties of networks employing memory properties [11] are better suited for 1D signals and pre-trained networks are bounded by the approach chosen during initial training which could be unoptimal for the task at hand.”
MethodBadry, E. A., Bedair, A., Atallah, H. A., & Essai, M. H. (2023) · SVU-International Journal of Engineering Sciences and Applications · doi:10.21608/svusrc.2023.177608.1088
Kanokoda, T., Kushitani, Y., Shimada, M., & Shirakashi, J. (2019) · Key engineering materials · doi:10.4028/www.scientific.net/kem.826.111
“Recently, several studies have shown that neural networks (NNs) can be useful methods in time series forecasting [7-11].”
BackgroundGrothendieck Category-Equivariant Neural Networks for Human Activity Recognition
Maruyama, Y., & Yasuda, A. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-29593-4_18 · Cited by 4
StitchSense: Smart-Garment Sensing as a Source Domain for Watch-Based Activity Recognition
Datta, I., Leng, Z., Bhayana, R., & Ploetz, T. (2026) · ACM International Symposium on Wearable Computers (ISWC) · doi:10.1145/3830727.3834837 · Cited by 1
A dissimilarity measure estimation for analyzing trajectory data
Arfa, R., Yusof, R. B., & Shabanzadeh, P. (2019) · Journal of Advanced Simulation in Science and Engineering · doi:10.15748/jasse.6.367
“One popular modification of RNN is LSTM [32] which has been successfully applied to problems in many domains [33-35] Bidirectional LSTM (BLSTM) is a variant of LSTM which has been validated to achieve state-of-the-art performance in many sequential modeling tasks including speech recognition [36],…”
MethodAnderson, H. E., Scheidt, R. A., & Bassindale, K. D. (2026) · Sensors · doi:10.3390/s26175357
Sheng, T., & Huber, M. (2025) · 2025 International Conference on Machine Learning and Applications (ICMLA) · doi:10.1109/icmla66185.2025.00049
“Methods ACC Probability SVM with Filter [3] 0.9678 DeepConvLSTM [6] 0.8942 Self-Attention [26] 0.8500 ResGCNN [27] 0.8313 Vanilla Autoencoder 0.6369 Proposed approach with 1% labels 0.8015 Proposed approach with 5% labels 0.8853 Proposed approach with 10% labels 0.9352 • The model outperforms the…”
InfluentialInstructor Activity Recognition Using Smartwatch and Smartphone Sensors
Chowdhury, Z. U. (2020)
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
“The baseline methods include standard LSTM [12], DeepConvLSTM [19], and LSTM-S (LSTM + sample-by-sample analysis) [11].”
BackgroundMethodInfluentialJacobsen, E. L., & Teizer, J. (2022) · Proceedings of the 29th EG-ICE International Workshop on Intelligent Computing in Engineering · doi:10.7146/aul.455.c232
Advanced AI Systems for Noise Suppression in Wireless Signal Transmission Using Deep Learning
Jyothi, M., Senthamarai, R., Narasimhulu, K., Kumar, G. R., Naidu, G. R., & Nathiya, M. (2025) · International Conference on Advancement in Futuristic Technologies (ICAFT) · doi:10.1109/icaft66710.2025.11453152
Swarm Intelligence for Data Management and Mining Technologies to Manage and Analyze Data in IoT
Chauhan, K., & Dutt, V. (2020) · Wiley eBooks · doi:10.1002/9781119778868.ch11
Chen, T., & Huo, G. (2026) · Journal of King Saud University - Computer and Information Sciences · doi:10.1007/s44443-026-00943-9 · Cited by 1
Taş, G., Kaymak, Ç., & Ölmez, Y. (2026) · Mühendislik Bilimleri ve Tasarım Dergisi · doi:10.21923/jesd.1814244
Yamamoto, T., Sawada, N., Sato, Y., Amesaka, T., & Sugiura, Y. (2025) · International Journal of Human-Computer Interaction · doi:10.1080/10447318.2025.2589497
“In this study, we focused on CNN–RNN hybrid architectures, which are widely used in IMU-based human activity recognition (HAR) tasks (Chen et al., 2021; Ord � o ~ nez & Roggen, 2016).”
Metadata
- Authors
- Ordóñez, F., & Roggen, D.
- Affiliations
- University of Sussex (GB)
- Corresponding author
- Ordóñez, Francisco
- Venue
- Sensors
- Publisher
- MDPI AG
- Published
- 2016-01-18
- Volume
- 16
- Issue
- 1
- Pages
- 115
- ISSN
- 1424-8220 (ISSN-L)
- PMC
- PMC4732148
- OpenAlex
- W2270470215
- Cited by
- 2,761
- References
- 49
- Citations by year
- 2026: 284 · 2025: 256 · 2024: 284 · 2023: 270 · 2022: 319 · 2021: 379 · 2020: 347 · 2019: 288 · 2018: 219 · 2017: 93 · 2016: 21
- FWCI
- 140.52
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- Locations
- Sensors (publishedVersion, cc-by)
- PubMed Central (publishedVersion, cc-by)
- PubMed (publishedVersion)
- Sussex Research Online (University of Sussex) (submittedVersion, cc-by)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- Figshare (submittedVersion, cc-by)
- Funders
- Google Faculty Research Award (Is deep learning useful for wearable activity recognition?)
- EPSRC (EP/N007816/1)
- Topics
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Context-Aware Activity Recognition Systems
- Physical Sciences › Engineering › Biomedical Engineering › Gait Recognition and Analysis
- Physical Sciences › Computer Science › Signal Processing › Time Series Analysis and Forecasting
- Keywords
- human activity recognition, wearable sensors, multimodal sensor fusion, convolutional neural networks, LSTM neural network, temporal dynamics, automatic feature extraction, hyperparameter optimization
- MeSH terms
- Machine Learning, Clothing, Humans, Human Activities, Signal Processing, Computer-Assisted, Databases, Factual, Neural Networks, Computer, Monitoring, Ambulatory
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
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.