OBD-II Driving Behavior Assessment: Analysis for Driving Styles Detection
Abstract
This paper presents a novel approach for driving style classification using exclusively On-Board Diagnostics (OBD-II) vehicle data. Key parameters including throttle position, revolutions per minute (RPM), and mass air flow (MAF), among others were collected via an ELM-327 adapter from a 2018 Honda HR-V along predefined urban routes driven by five participants.
Applied to a dataset of 7,349 OBD-II observations collected from five drivers, K-means clustering with elbow method validation and a Silhouette Score of 0.63 identified three behaviorally meaningful driving profiles: normal, moderate, and dynamic, with k=3 selected based on its interpretability and alignment with the experimental conditions. The results demonstrate effective behavioral segmentation using only internal vehicle parameters, eliminating the need for external sensors. This method proves particularly valuable for cost-sensitive applications, offering reliable driver behavior characterization through widely accessible telematics. Future work will leverage the generated clusters as ground-truth labels for supervised learning algorithms — including Support Vector Machines (SVM), Random Forest, and K-Nearest Neighbors (KNN) — to develop real-time classification systems.
Applied to a dataset of 7,349 OBD-II observations collected from five drivers, K-means clustering with elbow method validation and a Silhouette Score of 0.63 identified three behaviorally meaningful driving profiles: normal, moderate, and dynamic, with k=3 selected based on its interpretability and alignment with the experimental conditions. The results demonstrate effective behavioral segmentation using only internal vehicle parameters, eliminating the need for external sensors. This method proves particularly valuable for cost-sensitive applications, offering reliable driver behavior characterization through widely accessible telematics. Future work will leverage the generated clusters as ground-truth labels for supervised learning algorithms — including Support Vector Machines (SVM), Random Forest, and K-Nearest Neighbors (KNN) — to develop real-time classification systems.
Keywords
Driving styles, road safety, machine learning.