Nazish Ashfaq
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Nazish Ashfaq has a total of 17 co-authors. Here are the top co-authors by number of shared publications:
- Muhammad Hassan Khan
Faculty of Computing & Information Technology, University of the Punjab, Lahore, Pakistan
3 shared publications - Muhammad Adeel Nisar
Department of Information Technology, University of the Punjab, Lahore 54000, Pakistan
2 shared publications - Abeer Alubaidi
Department of Law, College of Shariah andLaw, Taif University, Taif 21944, Saudi Arabia
1 shared publication - Adnan Khalid
Faculty of Computing, Universiti Teknologi Malaysia, Johor Bahru, Malaysia
1 shared publication - Adnan Khalid
Department of Computer Science, Bahria University Lahore Campus, Lahore, Pakistan
1 shared publication - Asma Irshad
School of Biochemistry and Biotechnology, University of the Punjab, Lahore, Pakistan
1 shared publication
Latest Publications
- 9 June 2026
- Published by Springer Nature in The Journal of Supercomputing
- Vol. 82 (8) , 487
Ccnl-fallnet: an ensemble deep learning approach for fall detection with the humCareFall dataset
- 9 June 2026
- Published by Springer Nature in The Journal of Supercomputing
- Vol. 82 (8) , 487
Ccnl-fallnet: an ensemble deep learning approach for fall detection with the humCareFall dataset
- Open Access
- 9 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 104106-104119
HCMMA-Net: A Hybrid Convolutional Multi-Modal Attention Network for Human Activity Recognition in Smart Homes Using Wearable Sensor Data
Human activity recognition (HAR) plays a pivotal role in applications such as healthcare monitoring, fitness tracking, and smart homes. Multi-modal sensor data from wearable devices offers diverse perspectives on human motion, enhancing recognition accuracy and robustness. However, integrating these modalities poses challenges due to sensor heterogeneity and variability in placement. This study examines the role of multi-modalities in HAR using a hybrid convolutional multi-modal attention network (HCMMA-Net), designed to exploit spatial and temporal dependencies in sensor data. We evaluate the model on two benchmark datasets Cogage, achieving an accuracy of 93.94%, and WISDM, with an accuracy of 99.29%, demonstrating its strong generalizability across varied sensor configurations. Additionally, we present a newly collected multi-modal dataset, HumcareV1.0, comprising different activities in smart-homelike scenarios. On this real-world dataset, HCMMA-Net attains an accuracy of 97.56%, highlighting its effectiveness in capturing subtle behavioral nuances in practical environments. The model exhibits robust generalization across complex activity patterns and sensor configurations, underscoring the significance of multi-modal integration in advancing HAR systems. These findings highlight the potential of our approach for deployment in real-time, context-aware smart environments.
- 9 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 104106-104119
Open AccessHCMMA-Net: A Hybrid Convolutional Multi-Modal Attention Network for Human Activity Recognition in Smart Homes Using Wearable Sensor Data
Human activity recognition (HAR) plays a pivotal role in applications such as healthcare monitoring, fitness tracking, and smart homes. Multi-modal sensor data from wearable devices offers diverse perspectives on human motion, enhancing recognition accuracy and robustness. However, integrating these modalities poses challenges due to sensor heterogeneity and variability in placement. This study examines the role of multi-modalities in HAR using a hybrid convolutional multi-modal attention network (HCMMA-Net), designed to exploit spatial and temporal dependencies in sensor data. We evaluate the model on two benchmark datasets Cogage, achieving an accuracy of 93.94%, and WISDM, with an accuracy of 99.29%, demonstrating its strong generalizability across varied sensor configurations. Additionally, we present a newly collected multi-modal dataset, HumcareV1.0, comprising different activities in smart-homelike scenarios. On this real-world dataset, HCMMA-Net attains an accuracy of 97.56%, highlighting its effectiveness in capturing subtle behavioral nuances in practical environments. The model exhibits robust generalization across complex activity patterns and sensor configurations, underscoring the significance of multi-modal integration in advancing HAR systems. These findings highlight the potential of our approach for deployment in real-time, context-aware smart environments.
- Top Cited PapersOpen Access
- 4 May 2025
- Published by Springer Nature in Discover Computing
- Vol. 28 (1) , 1-18
SkinSight: advancing deep learning for skin cancer diagnosis and classification
- 4 May 2025
- Published by Springer Nature in Discover Computing
- Vol. 28 (1) , 1-18
Top Cited PapersOpen AccessSkinSight: advancing deep learning for skin cancer diagnosis and classification
- Open Access
- 8 March 2025
- Published by 50Sea in International Journal of Agriculture & Sustainable Development
- Vol. 7 (1) , 506-522
An Efficient and Robust Deep Learning Approach for Vehicle Recognition using Light-weight Deep Network
In the realm of intelligent transportation systems, automatic number plate detectionhas emerged as a crucial research topic due to its wide range of applications, including traffic violation monitoring, support for autonomous vehicles, vehicle speed tracking, automated toll collection, stolen vehicle identification, and overall traffic management. The goal of automatic number plate detection is to accurately identify vehicles based on their number plates. This study proposes a hierarchicalapproach for detecting number plates. In the initial phase, a lightweight deep learning model, Mobile Net-SSD, is employed to detect number plates. Subsequently, the alphanumeric characters from the detected number plates are extracted using an Optical Character Recognition (OCR) technique. The model is, built on a convolutional neural network, and efficiently uses depth wiseand pointwise convolution layers, making it suitable for mobile and embedded systems. Additionally, we introduce a dataset of 30,613 vehicle number plate images to foster further research. Experimental evaluations show that the proposed method achieves 95% accuracy on this dataset, significantly enhancing real-time number plate detection and making it suitable for large-scale implementations in smart cities and intelligent transportation networks.
- 8 March 2025
- Published by 50Sea in International Journal of Agriculture & Sustainable Development
- Vol. 7 (1) , 506-522
Open AccessAn Efficient and Robust Deep Learning Approach for Vehicle Recognition using Light-weight Deep Network
In the realm of intelligent transportation systems, automatic number plate detectionhas emerged as a crucial research topic due to its wide range of applications, including traffic violation monitoring, support for autonomous vehicles, vehicle speed tracking, automated toll collection, stolen vehicle identification, and overall traffic management. The goal of automatic number plate detection is to accurately identify vehicles based on their number plates. This study proposes a hierarchicalapproach for detecting number plates. In the initial phase, a lightweight deep learning model, Mobile Net-SSD, is employed to detect number plates. Subsequently, the alphanumeric characters from the detected number plates are extracted using an Optical Character Recognition (OCR) technique. The model is, built on a convolutional neural network, and efficiently uses depth wiseand pointwise convolution layers, making it suitable for mobile and embedded systems. Additionally, we introduce a dataset of 30,613 vehicle number plate images to foster further research. Experimental evaluations show that the proposed method achieves 95% accuracy on this dataset, significantly enhancing real-time number plate detection and making it suitable for large-scale implementations in smart cities and intelligent transportation networks.
- Open Access
- 11 June 2024
- Published by MDPI AG in Information
- Vol. 15 (6) , 343
Identification of Optimal Data Augmentation Techniques for Multimodal Time-Series Sensory Data: A Framework
Recently, the research community has shown significant interest in the continuous temporal data obtained from motion sensors in wearable devices. These data are useful for classifying and analysing different human activities in many application areas such as healthcare, sports and surveillance. The literature has presented a multitude of deep learning models that aim to derive a suitable feature representation from temporal sensory input. However, the presence of a substantial quantity of annotated training data is crucial to adequately train the deep networks. Nevertheless, the data originating from the wearable devices are vast but ineffective due to a lack of labels which hinders our ability to train the models with optimal efficiency. This phenomenon leads to the model experiencing overfitting. The contribution of the proposed research is twofold: firstly, it involves a systematic evaluation of fifteen different augmentation strategies to solve the inadequacy problem of labeled data which plays a critical role in the classification tasks. Secondly, it introduces an automatic feature-learning technique proposing a Multi-Branch Hybrid Conv-LSTM network to classify human activities of daily living using multimodal data of different wearable smart devices. The objective of this study is to introduce an ensemble deep model that effectively captures intricate patterns and interdependencies within temporal data. The term “ensemble model” pertains to fusion of distinct deep models, with the objective of leveraging their own strengths and capabilities to develop a solution that is more robust and efficient. A comprehensive assessment of ensemble models is conducted using data-augmentation techniques on two prominent benchmark datasets: CogAge and UniMiB-SHAR. The proposed network employs a range of data-augmentation methods to improve the accuracy of atomic and composite activities. This results in a 5% increase in accuracy for composite activities and a 30% increase for atomic activities.
- 11 June 2024
- Published by MDPI AG in Information
- Vol. 15 (6) , 343
Open AccessIdentification of Optimal Data Augmentation Techniques for Multimodal Time-Series Sensory Data: A Framework
Recently, the research community has shown significant interest in the continuous temporal data obtained from motion sensors in wearable devices. These data are useful for classifying and analysing different human activities in many application areas such as healthcare, sports and surveillance. The literature has presented a multitude of deep learning models that aim to derive a suitable feature representation from temporal sensory input. However, the presence of a substantial quantity of annotated training data is crucial to adequately train the deep networks. Nevertheless, the data originating from the wearable devices are vast but ineffective due to a lack of labels which hinders our ability to train the models with optimal efficiency. This phenomenon leads to the model experiencing overfitting. The contribution of the proposed research is twofold: firstly, it involves a systematic evaluation of fifteen different augmentation strategies to solve the inadequacy problem of labeled data which plays a critical role in the classification tasks. Secondly, it introduces an automatic feature-learning technique proposing a Multi-Branch Hybrid Conv-LSTM network to classify human activities of daily living using multimodal data of different wearable smart devices. The objective of this study is to introduce an ensemble deep model that effectively captures intricate patterns and interdependencies within temporal data. The term “ensemble model” pertains to fusion of distinct deep models, with the objective of leveraging their own strengths and capabilities to develop a solution that is more robust and efficient. A comprehensive assessment of ensemble models is conducted using data-augmentation techniques on two prominent benchmark datasets: CogAge and UniMiB-SHAR. The proposed network employs a range of data-augmentation methods to improve the accuracy of atomic and composite activities. This results in a 5% increase in accuracy for composite activities and a 30% increase for atomic activities.