A Framework for Combining Lateral and Longitudinal Acceleration to Assess Driving Styles Using Unsupervised Approach
DOI: 10.1109/tits.2023.3310213
archive: archived pipeline: cataloged verified
Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)
Summary
This paper addresses the challenge of assessing driving styles using large-scale, unlabeled naturalistic driving data, a task previously hindered by the reliance on supervised algorithms requiring manual labeling. The authors propose an unsupervised framework that combines lateral and longitudinal accelerations to evaluate driver risk and behavior. The motivation stems from the increasing availability of sensor data in intelligent transportation systems and the need for objective, scalable methods to support applications like usage-based insurance and driving feedback. The study specifically leverages the G-G diagram, which plots lateral against longitudinal acceleration, to capture the interdependence of these maneuvers, which previous studies often treated independently or neglected. The methodology employs a two-stage clustering approach applied to a dataset of 830,338 km of driving data from 71 elderly drivers collected via in-vehicle recorders. First, a "safe driving area" is defined statistically based on the distribution of all drivers' data to identify risky acceleration maneuvers. Hierarchical clustering then classifies drivers into high-, medium-, and low-risk groups based on the frequency of these risky maneuvers and a driving instability index. Second, a unique Gaussian Mixture Model (GMM) is trained for each individual driver to decompose their risky maneuvers into specific risk components. This allows for the calculation of a normalized driving performance score (0–100) based on the probability and severity of these patterns. Finally, the framework extracts spatio-temporal characteristics, such as behavioral hotspots, to enhance the interpretability of the machine learning results. The results demonstrate that combining lateral and longitudinal accelerations is necessary for accurate driver behavior assessment, as the two axes are correlated. The proposed unsupervised method effectively models individual driving styles from the G-G diagram without requiring labeled data. The validity of the calculated performance scores was confirmed using drivers' self-reported crash data. The framework successfully generates comprehensive risk profiles for each driver, including their risk level, performance score, specific risk components, and spatio-temporal context of risky behaviors. The significance of this work lies in its ability to process large-scale unlabeled data, addressing a major limitation in current driving style assessment literature. By using unsupervised learning, the method avoids subjective labeling and manual encoding of G-G diagrams. Furthermore, the inclusion of spatio-temporal context improves the interpretability of the results, providing actionable insights for drivers to avoid risky circumstances. This framework enables the creation of tailored feedback and accurate risk profiling for intelligent transportation applications, promoting safer and more efficient driving behaviors.
Provenance
The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.
| Stage | Outcome | Tool | Model | Prompt | Attempts | Completed |
|---|---|---|---|---|---|---|
| discover | success | OpenAlex-citations | — | — | 1 | 2026-06-20 |
| archive | success | unpaywall | — | — | 2 | 2026-06-26 |
| extract | success | cached | — | — | 2 | 2026-06-26 |
| clean | success | clean | — | — | 1 | 2026-06-20 |
| chunk | success | chunk | — | — | 1 | 2026-06-20 |
| embed | success | embed | Qwen/Qwen3-Embedding-8B | — | 1 | 2026-06-20 |
| promote | success | — | — | — | 1 | 2026-06-20 |
| summarize | success | llm | qwen3.6-27b-prismaquant | summ-v5 | 1 | 2026-06-26 |
| tag | success | vector_similarity | — | — | 6 | 2026-06-20 |
| verify | success | — | — | — | 1 | 2026-06-26 |
Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.
Topics
Ranked by relevance to this paper. Hover a topic for its definition.
Information type
What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).
- Methodological Resource: tool software, dataset resource
- Theoretical Contribution: computational model