ITSC 2025 Paper Abstract

Paper TH-EA-T29.5

Styanidis, Aristotelis (National Technical University of Athens, CardioID Technologies), Lourenço, André (CardioID Technologies), Carreiras, Carlos (CardioID Technologies Lda), Ahlström, Christer (Swedish National Road and Transport Research Institute), Ziakopoulos, Apostolos (National Technical University of Athens), Yannis, George (National Technical University of Athens)

Impact of Missing Inter-Beat Intervals on Heart Rate Variability Features and Driver Drowsiness Detection

Scheduled for presentation during the Regular Session "S29b-Human Factors and Human Machine Interaction in Automated Driving" (TH-EA-T29), Thursday, November 20, 2025, 14:50−14:50, Currumbin

2025 IEEE 28th International Conference on Intelligent Transportation Systems (ITSC), November 18-21, 2025, Gold Coast, Australia

This information is tentative and subject to change. Compiled on October 18, 2025

Keywords Driver Behavior Monitoring and Feedback Systems for Semi-autonomous Vehicles, Human-Machine Interaction Systems for Enhanced Driver Assistance and Safety

Abstract

Drowsiness detection using physiological data, particularly heart rate variability (HRV), has emerged as a promising approach for assessing driver fatigue. Inter-beat intervals (IBIs) represent the time between consecutive heartbeats and are used to derive HRV, a measure of autonomic nervous system activity. HRV can be extracted from electrocardiogram (ECG), photoplethysmography (PPG) or other physiological signals, making it valuable for applications in health monitoring and driver state assessment. However, real-world conditions often lead to missing or corrupted data due to inconsistent sensor contact, motion artifacts, or signal interruptions. Such data loss can impact HRV feature extraction and affect downstream machine learning (ML) models. This study investigates the effects of missing data by systematically removing 15–30% of IBIs, either randomly or in sequential blocks. The distribution analysis indicated that IBI distributions largely retain their overall structure, with only minor deviations. Classification performance was robust to the investigated data losses, but when comparing 10-fold cross-validation to leave-one-participant-out validation, mean accuracy dropped by approximately 15%, and variability in accuracy across folds increased by around 10%. tSNE feature space visualization further revealed that class separability was much clearer at the participant level than at the group level. The findings underscore the need for personalized models tailored to each driver's physiological patterns, which has implications for drowsiness detection systems.