Author Identifier
Date of Award
2026
Keywords
personalised machine learning, idiographic modelling, multimodal physiological signals, EEG, galvanic skin response, driver distraction, cognitive load, explainable AI
Document Type
Thesis
Publisher
Edith Cowan University
Degree Name
Doctor of Philosophy
School
School of Arts and Humanities
First Supervisor
Craig Speelman
Second Supervisor
Mike Johnstone
Abstract
Cognitive load during driving refers to the mental effort required to manage traffic demands while maintaining situational awareness. Elevated cognitive load can impair driver performance and increase the risk of road traffic accidents, yet detecting cognitive distraction remains challenging because it often occurs without observable behavioural cues. Physiological signals provide a promising means of capturing internal cognitive states; however, most existing driver monitoring approaches rely on group-level machine learning models that assume cognitive load is expressed uniformly across individuals. Growing evidence suggests that physiological responses to cognitive demand vary substantially between drivers, raising questions about the effectiveness of group-level modelling strategies.
This thesis investigates whether cognitive load during driving can be detected using multimodal physiological signals and whether personalised modelling approaches improve detection performance compared to group-level modelling. Fifty licensed drivers participated in controlled simulated driving experiments involving two conditions: normal driving and cognitively demanding dual-task driving involving continuous mental arithmetic. Electroencephalography (EEG), galvanic skin response (GSR), and heart rate (HR) signals were recorded continuously during the driving sessions. Multiple machine learning architectures, including classical classifiers and deep learning models, were implemented under two modelling paradigms: group-level models trained across participants and personalised within-subject models trained for individual drivers. Model performance was evaluated using window-based analysis and standard classification metrics.
The results demonstrate that cognitive load can be reliably distinguished from normal driving using physiological signals. EEG provided the most discriminative information, while GSR and heart rate contributed complementary autonomic indicators of cognitive demand. However, substantial inter-individual variability was observed in physiological response patterns. Across all evaluated algorithms, personalised models consistently outperformed group-level models at the participant level, whereas group-level models exhibited limited generalisation when applied to unseen drivers.
These findings indicate that physiological expressions of cognitive load during driving are non-uniform across individuals and that personalised modelling provides a more effective framework for physiological driver monitoring systems. The thesis contributes empirical evidence supporting personalised machine learning approaches for cognitive load detection and highlights the importance of individual-specific calibration in the development of adaptive driver monitoring technologies for intelligent transport systems. Feature attribution analysis using SHAP (SHapley Additive exPlanations) further revealed EEG as the dominant physiological modality contributing to model predictions, with GSR providing complementary contributions.
Access Note
Access to this thesis is embargoed until 12th August 2028
Recommended Citation
Aponso, G. (2026). Personalised machine learning for driver distraction prediction. Edith Cowan University. https://doi.org/10.25958/h1sj-zk45