HybridHAR-Net: Recognizing Human Activities Using a Hybrid Deep Learning-Based Model for Mobile Health Applications

Debarshi Bhattacharya; Karam Kumar Sahoo; Pawan Kumar Singh; Mufti Mahmud

doi:10.1007/978-981-96-3860-4_12

HybridHAR-Net: Recognizing Human Activities Using a Hybrid Deep Learning-Based Model for Mobile Health Applications

Debarshi Bhattacharya
, Karam Kumar Sahoo
, Pawan Kumar Singh^*
, Mufti Mahmud

^*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

Abstract

The classification of human motions into their corresponding activity classes is an integral challenge for Human Activity Recognition (HAR) since the data is available in multiple modalities; hence, a single framework or approach is not enough to handle this. In this paper, we have applied Continuous Wavelet Transformation (CWT) to convert the raw time-series sensor data into 3D matrices of n-channels analogous to multi-channel images. These matrices are fed into our proposed hybrid model named HybridHAR-Net, which comprises a two-dimensional Convolutional Neural Network (CNN) architecture and a multi-layered Gated Recurrent Unit (GRU) network. The CNN architecture of HybridHAR-Net extracts structural features from the n-channel inputs while the layered GRU network processes the sequential features, producing a better HAR evaluation result. The proposed HybridHAR-Net model is evaluated on two publicly available benchmark datasets- WISDM and UCI-HAR and achieved classification accuracies of 97.43% and 97.15%, respectively. The experiment results show that our proposed method of utilizing CWT on raw sensor data and applying the proposed HybridHAR-Net model for predicting daily-life human activities outperforms recent multiple HAR methods to which our method is compared.

Original language	English
Title of host publication	Intelligent Computing Systems and Applications - Select Proceedings of the International Conference, ICICSA 2022
Editors	Saroj Kumar Biswas, Sivaji Bandyopadhyay, Yoichi Hayashi, Valentina Emilia Balas
Publisher	Springer Science and Business Media Deutschland GmbH
Pages	145-157
Number of pages	13
ISBN (Print)	9789819638598
DOIs	https://doi.org/10.1007/978-981-96-3860-4_12
State	Published - 2025
Externally published	Yes
Event	1st International Conference on Intelligent Computing Systems and Applications, ICICSA 2022 - Silchar, India Duration: 21 Sep 2023 → 22 Sep 2023

Publication series

Name	Lecture Notes in Networks and Systems
Volume	1307 LNNS
ISSN (Print)	2367-3370
ISSN (Electronic)	2367-3389

Conference

Conference	1st International Conference on Intelligent Computing Systems and Applications, ICICSA 2022
Country/Territory	India
City	Silchar
Period	21/09/23 → 22/09/23

Bibliographical note

Publisher Copyright:
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.

Keywords

Continuous wavelet transformation
Human activity recognition
Hybrid deep learning model
HybridHAR-Net
UCI-HAR
WISDM

ASJC Scopus subject areas

Control and Systems Engineering
Signal Processing
Computer Networks and Communications

Access to Document

10.1007/978-981-96-3860-4_12

Cite this

Bhattacharya, D., Sahoo, K. K., Singh, P. K., & Mahmud, M. (2025). HybridHAR-Net: Recognizing Human Activities Using a Hybrid Deep Learning-Based Model for Mobile Health Applications. In S. K. Biswas, S. Bandyopadhyay, Y. Hayashi, & V. E. Balas (Eds.), Intelligent Computing Systems and Applications - Select Proceedings of the International Conference, ICICSA 2022 (pp. 145-157). (Lecture Notes in Networks and Systems; Vol. 1307 LNNS). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-981-96-3860-4_12

Bhattacharya, Debarshi ; Sahoo, Karam Kumar ; Singh, Pawan Kumar et al. / HybridHAR-Net : Recognizing Human Activities Using a Hybrid Deep Learning-Based Model for Mobile Health Applications. Intelligent Computing Systems and Applications - Select Proceedings of the International Conference, ICICSA 2022. editor / Saroj Kumar Biswas ; Sivaji Bandyopadhyay ; Yoichi Hayashi ; Valentina Emilia Balas. Springer Science and Business Media Deutschland GmbH, 2025. pp. 145-157 (Lecture Notes in Networks and Systems).

@inproceedings{52d38e9274964eae9b316589365fdbc7,

title = "HybridHAR-Net: Recognizing Human Activities Using a Hybrid Deep Learning-Based Model for Mobile Health Applications",

abstract = "The classification of human motions into their corresponding activity classes is an integral challenge for Human Activity Recognition (HAR) since the data is available in multiple modalities; hence, a single framework or approach is not enough to handle this. In this paper, we have applied Continuous Wavelet Transformation (CWT) to convert the raw time-series sensor data into 3D matrices of n-channels analogous to multi-channel images. These matrices are fed into our proposed hybrid model named HybridHAR-Net, which comprises a two-dimensional Convolutional Neural Network (CNN) architecture and a multi-layered Gated Recurrent Unit (GRU) network. The CNN architecture of HybridHAR-Net extracts structural features from the n-channel inputs while the layered GRU network processes the sequential features, producing a better HAR evaluation result. The proposed HybridHAR-Net model is evaluated on two publicly available benchmark datasets- WISDM and UCI-HAR and achieved classification accuracies of 97.43\% and 97.15\%, respectively. The experiment results show that our proposed method of utilizing CWT on raw sensor data and applying the proposed HybridHAR-Net model for predicting daily-life human activities outperforms recent multiple HAR methods to which our method is compared.",

keywords = "Continuous wavelet transformation, Human activity recognition, Hybrid deep learning model, HybridHAR-Net, UCI-HAR, WISDM",

author = "Debarshi Bhattacharya and Sahoo, \{Karam Kumar\} and Singh, \{Pawan Kumar\} and Mufti Mahmud",

note = "Publisher Copyright: {\textcopyright} The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.; 1st International Conference on Intelligent Computing Systems and Applications, ICICSA 2022 ; Conference date: 21-09-2023 Through 22-09-2023",

year = "2025",

doi = "10.1007/978-981-96-3860-4\_12",

language = "English",

isbn = "9789819638598",

series = "Lecture Notes in Networks and Systems",

publisher = "Springer Science and Business Media Deutschland GmbH",

pages = "145--157",

editor = "Biswas, \{Saroj Kumar\} and Sivaji Bandyopadhyay and Yoichi Hayashi and Balas, \{Valentina Emilia\}",

booktitle = "Intelligent Computing Systems and Applications - Select Proceedings of the International Conference, ICICSA 2022",

address = "Germany",

}

Bhattacharya, D, Sahoo, KK, Singh, PK & Mahmud, M 2025, HybridHAR-Net: Recognizing Human Activities Using a Hybrid Deep Learning-Based Model for Mobile Health Applications. in SK Biswas, S Bandyopadhyay, Y Hayashi & VE Balas (eds), Intelligent Computing Systems and Applications - Select Proceedings of the International Conference, ICICSA 2022. Lecture Notes in Networks and Systems, vol. 1307 LNNS, Springer Science and Business Media Deutschland GmbH, pp. 145-157, 1st International Conference on Intelligent Computing Systems and Applications, ICICSA 2022, Silchar, India, 21/09/23. https://doi.org/10.1007/978-981-96-3860-4_12

HybridHAR-Net: Recognizing Human Activities Using a Hybrid Deep Learning-Based Model for Mobile Health Applications. / Bhattacharya, Debarshi; Sahoo, Karam Kumar; Singh, Pawan Kumar et al.
Intelligent Computing Systems and Applications - Select Proceedings of the International Conference, ICICSA 2022. ed. / Saroj Kumar Biswas; Sivaji Bandyopadhyay; Yoichi Hayashi; Valentina Emilia Balas. Springer Science and Business Media Deutschland GmbH, 2025. p. 145-157 (Lecture Notes in Networks and Systems; Vol. 1307 LNNS).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - HybridHAR-Net

T2 - 1st International Conference on Intelligent Computing Systems and Applications, ICICSA 2022

AU - Bhattacharya, Debarshi

AU - Sahoo, Karam Kumar

AU - Singh, Pawan Kumar

AU - Mahmud, Mufti

N1 - Publisher Copyright: © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.

PY - 2025

Y1 - 2025

N2 - The classification of human motions into their corresponding activity classes is an integral challenge for Human Activity Recognition (HAR) since the data is available in multiple modalities; hence, a single framework or approach is not enough to handle this. In this paper, we have applied Continuous Wavelet Transformation (CWT) to convert the raw time-series sensor data into 3D matrices of n-channels analogous to multi-channel images. These matrices are fed into our proposed hybrid model named HybridHAR-Net, which comprises a two-dimensional Convolutional Neural Network (CNN) architecture and a multi-layered Gated Recurrent Unit (GRU) network. The CNN architecture of HybridHAR-Net extracts structural features from the n-channel inputs while the layered GRU network processes the sequential features, producing a better HAR evaluation result. The proposed HybridHAR-Net model is evaluated on two publicly available benchmark datasets- WISDM and UCI-HAR and achieved classification accuracies of 97.43% and 97.15%, respectively. The experiment results show that our proposed method of utilizing CWT on raw sensor data and applying the proposed HybridHAR-Net model for predicting daily-life human activities outperforms recent multiple HAR methods to which our method is compared.

AB - The classification of human motions into their corresponding activity classes is an integral challenge for Human Activity Recognition (HAR) since the data is available in multiple modalities; hence, a single framework or approach is not enough to handle this. In this paper, we have applied Continuous Wavelet Transformation (CWT) to convert the raw time-series sensor data into 3D matrices of n-channels analogous to multi-channel images. These matrices are fed into our proposed hybrid model named HybridHAR-Net, which comprises a two-dimensional Convolutional Neural Network (CNN) architecture and a multi-layered Gated Recurrent Unit (GRU) network. The CNN architecture of HybridHAR-Net extracts structural features from the n-channel inputs while the layered GRU network processes the sequential features, producing a better HAR evaluation result. The proposed HybridHAR-Net model is evaluated on two publicly available benchmark datasets- WISDM and UCI-HAR and achieved classification accuracies of 97.43% and 97.15%, respectively. The experiment results show that our proposed method of utilizing CWT on raw sensor data and applying the proposed HybridHAR-Net model for predicting daily-life human activities outperforms recent multiple HAR methods to which our method is compared.

KW - Continuous wavelet transformation

KW - Human activity recognition

KW - Hybrid deep learning model

KW - HybridHAR-Net

KW - UCI-HAR

KW - WISDM

UR - https://www.scopus.com/pages/publications/105018795610

U2 - 10.1007/978-981-96-3860-4_12

DO - 10.1007/978-981-96-3860-4_12

M3 - Conference contribution

AN - SCOPUS:105018795610

SN - 9789819638598

T3 - Lecture Notes in Networks and Systems

SP - 145

EP - 157

BT - Intelligent Computing Systems and Applications - Select Proceedings of the International Conference, ICICSA 2022

A2 - Biswas, Saroj Kumar

A2 - Bandyopadhyay, Sivaji

A2 - Hayashi, Yoichi

A2 - Balas, Valentina Emilia

PB - Springer Science and Business Media Deutschland GmbH

Y2 - 21 September 2023 through 22 September 2023

ER -

Bhattacharya D, Sahoo KK, Singh PK, Mahmud M. HybridHAR-Net: Recognizing Human Activities Using a Hybrid Deep Learning-Based Model for Mobile Health Applications. In Biswas SK, Bandyopadhyay S, Hayashi Y, Balas VE, editors, Intelligent Computing Systems and Applications - Select Proceedings of the International Conference, ICICSA 2022. Springer Science and Business Media Deutschland GmbH. 2025. p. 145-157. (Lecture Notes in Networks and Systems). doi: 10.1007/978-981-96-3860-4_12

HybridHAR-Net: Recognizing Human Activities Using a Hybrid Deep Learning-Based Model for Mobile Health Applications

Abstract

Publication series

Conference

Bibliographical note

Keywords

ASJC Scopus subject areas

Access to Document

Fingerprint

Cite this