Development of a statistical method for predicting human driver decisions.

Tan, Yaoyan Vincent; Elliott, Michael; Flannagan, Carol · 2015 · ROSA P / Center for Advancing Transportation Leadership and Safety (ATLAS Center)

archive: archived pipeline: cataloged verified

Get this paper ↗ (full text — opens at the source; we link to it, we don't host it)

Summary

This study addresses the challenge of enabling Level 4 autonomous vehicles to safely interact with human drivers, who often fail to clearly communicate their intentions. The authors hypothesize that the kinematic behavior of a human-driven vehicle, specifically its speed profile, can serve as a reliable predictor of driver intent within a short timeframe. The research focuses on predicting whether a human driver will stop before executing a left turn at an intersection, aiming to develop a statistical model that autonomous vehicles can use to anticipate such decisions and avoid collisions or unnecessary stops. The researchers utilized naturalistic driving data from the Integrated Vehicle-Based Safety Systems (IVBSS) study, involving 108 licensed drivers in Michigan. The dataset comprised 1,823 left turns recorded during baseline unsupervised driving periods. To facilitate prediction, time-series speed data was converted into a distance series relative to the intersection center, ranging from -100 meters to -1 meter. A distance-varying outcome was defined to indicate whether the vehicle would stop in the future at each specific meter. The methodology employed a moving window of recent speeds to capture relevant intent signals while minimizing noise. Principal Components Analysis (PCA) was used to reduce the dimensionality of the speed data, with the first three principal components explaining over 99% of the variation. These components were then linked to the stopping outcome using Bayesian Additive Regression Trees (BART). Model performance was evaluated using the Area Under the Receiver Operating Characteristic Curve (AUC), Capture Ratio (CR), and False Positive Ratio (FPR). The results demonstrated that the BART model, using a six-meter moving window and the first three principal components, achieved strong predictive performance. The AUC profile remained above 0.7 throughout the approach, starting at 0.75 at -95 meters and steadily increasing to over 0.90 by -25 meters from the intersection center, eventually reaching 1.0 as the vehicle neared the intersection. The first principal component corresponded to average speed, while the second resembled acceleration or deceleration. The analysis of CR and FPR profiles indicated that optimal probability cutoffs could be adjusted based on distance to balance the risk of unnecessary autonomous stops against the risk of crashes. The study concludes that kinematic speed data alone provides sufficient information to predict human driver decisions with high accuracy, offering a promising foundation for autonomous vehicle algorithms. The authors note that while the current model performs well, future improvements could include additional covariates such as the presence of lead vehicles or turn signal activation, particularly to enhance prediction accuracy at greater distances. They also recommend addressing intra-driver correlation through random intercept models and establishing feedback loops to simulate how autonomous vehicle behavior might influence human drivers. This work supports the development of safer interaction protocols between automated and human-driven vehicles.

Key finding

The prediction model achieved an area under the receiver operating characteristic curve of more than 0.90 at 25 meters away from the center of an intersection.

Methodology

naturalistic

Sample size: 108

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 bulk_ingest_rosap on 2026-05-23 (6 acquisition events logged).

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.

• telematics crash prediction
• anticipation

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).

• Empirical Findings: behavioral performance data
• Methodological Resource: dataset resource
• Theoretical Contribution: computational model