Fusion of ECG, PPG, and GSR Using Attention Based Machine Learning for Mental Workload Assessment

View at publisher Summarise with VIRAT
Authors: Zafar Khan, Sapna Prabhu
Cited by: 0
References: 23

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

An accurate and interpretable evaluation of mental workload (MW) is essential for understanding cognitive demands in stressful conditions for example driving, the aviation industry, and complex multitasking.This research introduces a multimodal physiological framework that integrates electrocardiogram, photoplethysmogram, and galvanic skin response signals obtained during N-back cognitive tasks to objectively categorize workload levels. We extracted time, frequency, and nonlinear-domain features after carefully preprocessing the data collected from 22 participants in the MOUS dataset. We evaluated three machine learning (ML) algorithms, Support Vector Machine (SVM), Random Forest (RF), and Gradient Boosting (GB), using nested cross-validation both before and after hyperparameter optimization.The results indicated that GB attained the highest overall performance (accuracy = 0.542, AUC = 0.563, F1 = 0.538), outperforming both RF and SVM. PPG was the most discriminative modality, while multimodal fusion (ECG + PPG + GSR) made the biggest overall improvement (AUC Δ + 0.054). An attention-based fusion mechanism also showed that PPG was more relevant when the workload was low, whereas ECG was more relevant when the mental workload was high. Overall, combining ensemble learning with hyperparameter tuning enhances accuracy and provides a simple and effective method for real-time cognitive workload monitoring.

Fields of Study & Subjects

Emotion and Mood Recognition Non-Invasive Vital Sign Monitoring Human-Automation Interaction and Safety

Publication details

Journal / source
Biennial International Conference on Nascent Technologies in Engineering (ICNTE)
Pages
1–6
Publication date
16 January 2026
Language
EN