An Adaptive Class-Wise Leader–Follower Pair Identification Framework and Behavioural Regimes in Heterogeneous Traffic
IEEE Access · 2026 · 14 · 120797-120828
DOI: 10.1109/access.2026.3712749 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Susan Eldhose und 2 weiteren Autor:innen. Die Arbeit nennt 57 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Women Leading IITM (WLI) initiative, Ministry of Human Resource Development, India, Half-Time Research Assistantship, Maha Kumbh Mela Study.
Abstract (Originalsprache)
Road traffic in developing countries operates under heterogeneous, weak lane disciplined (HWLD) conditions, with diverse vehicle types, and high manoeuvrability, making reliable identification of vehicle-following (VF) behaviour and leader–follower (LF) pairs challenging. Conventional proximity-based methods using fixed distance thresholds neither account for vehicle-type- and speed-dependent interaction distances nor verify dynamic coupling between LF speed profiles. Consequently, transient interactions are frequently misclassified as valid LF pairs, leading to biased behavioral inference and degraded model calibration. This study proposes a behaviour-consistent LF identification framework, quantifies reaction-time (RT) heterogeneity across vehicle classes and speed ranges, and captures within-pair behavioural variability. First, LF identification is refined using (i) vehicle-type- and speed-specific desirable gap thresholds derived from steady state flow–density diagrams, replacing the fixed distance criteria of prior methods, (ii) Wavelet Transform (WT) based LF speed profiles matching to verify dynamic coupling between leader and follower, a step absent in conventional identification methods, and (iii) kinematic indicators including relative-velocity sign-change ratio and longitudinal gap range ( $G_{k}$ ) grounded in classical car-following theory, to detect approaching and diverging interactions. These criteria systematically filter overtaking, tailgating, and unstable pairs, which proximity-based filters alone cannot distinguish, to construct behaviourally consistent LF datasets. Second, RT is estimated through calibration over candidate lags (0.5–2.5 s), combined with WT-based stimulus-response peak matching, and speed-stratified RT distributions. Third, clustering on kinematic features reveals distinct following regimes within each LF pair type. Compared with conventional unfiltered baselines, behaviour-consistent filtering improves model fit by 4–41% in adjusted $R^{2}$ across LF pairs in the primary dataset, with approaching/diverging filter eliminating most spurious pairs. To assess the generalisability of the framework, the filtering procedure was further applied to CAR–CAR pairs from two additional HWLD datasets. The $G_{k}$ threshold for the approaching/diverging filter remained consistent across all three datasets, ranging between 10.60 m and 10.82 m, while the overtaking speed threshold varied with the operating speed regime at each site, confirming that the filtering criteria are structurally transferable while numerical thresholds require site-specific recalibration. Results from the primary dataset further show that RT varies systematically with LF pair type and speed range rather than following a single representative value. Cluster-wise models further improve performance by 20–48% in adjusted $R^{2}$ compared with aggregate models, particularly for two-wheeler-involving pairs, demonstrating the importance of accounting for behavioural heterogeneity in VF modelling under HWLD mixed traffic.
Förderung
- Women Leading IITM (WLI) initiative
- Ministry of Human Resource Development, India, Half-Time Research Assistantship
- Maha Kumbh Mela Study
Zitieren
APA
Eldhose, S., Chilukuri, B. R., & Rajendran, C. (2026). An Adaptive Class-Wise Leader–Follower Pair Identification Framework and Behavioural Regimes in Heterogeneous Traffic. IEEE Access, 14, 120797-120828. https://doi.org/10.1109/access.2026.3712749
Vancouver
Eldhose S, Chilukuri BR, Rajendran C. An Adaptive Class-Wise Leader–Follower Pair Identification Framework and Behavioural Regimes in Heterogeneous Traffic. IEEE Access. 2026;14:120797-120828. doi:10.1109/access.2026.3712749
BibTeX
@article{Eldhose2026,
author = {Eldhose, Susan and Chilukuri, Bhargava Rama and Rajendran, Chandrasekharan},
title = {An Adaptive Class-Wise Leader–Follower Pair Identification Framework and Behavioural Regimes in Heterogeneous Traffic},
journal = {IEEE Access},
year = {2026},
volume = {14},
pages = {120797--120828},
doi = {10.1109/access.2026.3712749}
}
EUVIMED