Anomalous Behavior Detection in Trajectory Data of Older Drivers
DOI: 10.1109/honet59747.2023.10374878
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 detecting anomalous driving behaviors in older adults, a task critical for identifying Mild Cognitive Impairment (MCI) and ensuring driver safety. Older drivers often exhibit deviations such as missing exits, incorrect turns, or erratic speed changes due to reduced spatial awareness. The authors define the Anomalous Behavior Detection (ABD) problem as identifying trips with significant directional deviations, hard braking, and hard acceleration. The research is motivated by the computational difficulty of analyzing large, temporally detailed trajectory datasets and the need to distinguish between benign deviations (e.g., traffic avoidance) and abnormal patterns indicative of cognitive decline. The proposed methodology utilizes an Edge-Attributed Matrix to represent trip data within a road network. Each trip is modeled as a directed graph where nodes represent spatial locations and edges represent road segments. These edges are attributed with temporally detailed driving metrics, including average speed, direction, segment length, and counts of hard accelerations and brakes. The authors preprocess raw GPS and Inertial Measurement Unit (IMU) data by projecting points onto the nearest road segments. To identify anomalies, they employ the Isolation Forest algorithm, an unsupervised machine learning technique that isolates rare data points by constructing binary trees. This approach allows for the evaluation of trips with varying start and end points, assigning an abnormality score based on deviations from expected norms in speed and path geometry. Experiments were conducted using a real-world dataset collected from 18 senior citizens aged 65 to 85 over a three-month period in 2022. The data was recorded via in-vehicle telematics sensors capturing GPS coordinates and IMU dynamics. The study evaluated the model’s performance by classifying drivers as normal or abnormal based on the top 10% and 20% of anomaly scores, with a contamination parameter set to 0.2. Results indicated that the approach effectively identified abnormal driving behaviors. When considering the top 10% of drivers, the model achieved an accuracy of 0.83, though the F1-score was 0.40. For the top 20% threshold, accuracy decreased to 0.77, but the F1-score improved to 0.50, suggesting a better balance between precision and recall when a broader group of potential outliers is considered. The significance of this work lies in providing a baseline framework for monitoring cognitive health through driving behavior. By leveraging edge-attributed matrices and unsupervised learning, the method can detect subtle signs of impairment, such as cyclic routes or unexplained detours, without requiring labeled trip data. The authors conclude that while the current approach demonstrates promise, future work should expand the dataset to include more subjects and develop novel algorithms tailored specifically to the unique characteristics of senior drivers. This research contributes to the fields of trajectory data mining and spatio-temporal network analysis by offering a scalable solution for automated behavioral monitoring in aging populations.
Key finding
The proposed edge-attributed matrix and Isolation Forest approach successfully identified abnormal driving behaviors in older drivers, achieving an accuracy of 0.83 when classifying the top 10% of drivers with the highest anomaly scores.
Methodology
dataset
Sample size: 18
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. Discovered via author_sweep_intake on 2026-05-27.
| Stage | Outcome | Tool | Model | Prompt | Attempts | Completed |
|---|---|---|---|---|---|---|
| discover | success | author_sweep | — | — | 2 | 2026-05-27 |
| archive | success | canonical_url | — | — | 11 | 2026-06-06 |
| extract | success | cached | — | — | 3 | 2026-06-10 |
| clean | success | clean | — | — | 1 | 2026-06-04 |
| chunk | success | chunk | — | — | 1 | 2026-06-04 |
| embed | success | embed | Qwen/Qwen3-Embedding-8B | — | 1 | 2026-06-04 |
| enrich | success | semantic_scholar | — | — | 2 | 2026-06-04 |
| promote | success | — | — | — | 1 | 2026-06-04 |
| summarize | success | llm | qwen3.6-27b-prismaquant | summ-v5 | 2 | 2026-06-10 |
| tag | success | vector_similarity | — | — | 15 | 2026-06-11 |
| verify | success | — | — | — | 2 | 2026-06-10 |
Summary generated by qwen3.6-27b-prismaquant on 2026-06-10; 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: dataset resource
- Theoretical Contribution: computational model