SCIENCE@home

Dual Attention Network for Multimodal Physiological Signal-Based Fatigue Detection

Santosh Kumar,Sanchita Paul

Published 2025 in 2025 OITS International Conference on Information Technology (OCIT)

ABSTRACT

Maintaining focused attention and mental wellbeing is essential for safety in industrial environments, particularly for drivers operating under high-pressure and fastmoving conditions. This study proposes a Dual Attention Network (DAN) integrates temporal attention to capture critical time-dependent patterns and modality attention to dynamically weigh the contribution of EEG, ECG, GSR, and body temperature while also considering the driving stressor factors such as noise from traffic, overall temperature and humidity. This dual mechanism enhances interpretability and improves fatigue classification accuracy. For robust evaluation, the study uses two different datasets including the publicly available Fatigue Set dataset and locally collected multimodal dataset using wearable device for Indian Industrial drivers. This integration improved the diversity of subjects and driving conditions, enabling the DAN model to achieve higher generalizability. Demographic analysis revealed significant differences in fatigue across socio-economic status, marital status, geographical location, and age. Sleep duration showed a strong protective effect, inversely correlated with fatigue. Ground-truth labels were obtained using the Multidimensional Fatigue Inventory (MFI-20), and features including heart rate variability, EEG band powers, skin conductance, and thermal statistics were extracted. Multiple classifiers were trained after preprocessing and balancing (SMOTE). Among traditional approaches, Support Vector Machine (SVM) achieved the highest accuracy (90.28%), with Logistic Regression, and Gradient Boosting, RNN-LSTM hybrid model performing competitively. A Dual Attention Network (DAN) model, incorporating temporal and modality attention, further advanced performance, reaching 98.40 % accuracy. This research provides technical support for the development of affordable and dependable wearable motion monitoring devices.

PUBLICATION RECORD

CITATION MAP

EXTRACTION MAP

CLAIMS

  • No claims are published for this paper.

CONCEPTS

  • No concepts are published for this paper.

REFERENCES

Showing 1-15 of 15 references · Page 1 of 1

CITED BY

  • No citing papers are available for this paper.

Showing 0-0 of 0 citing papers · Page 1 of 1