MIMIC — Multidisciplinary Initiative on Methods to Integrate and Create Realistic Artificial Data

Edara, Praveen K.; Sun, Carlos; Brown, Henry; Savolainen, Peter T.; Shankar, Venky; Balakrishnan, Bimal; Shang, Yi; Chakraborty, Sounak; Adu-Gyamfi, Yaw; Li, Can; Aati, Khaled; Lima, S.; Huang, Yilun; Mussah, Abdul Rashid; Hopfenblatt, J. · 2023 · ROSA P / United States. Federal Highway Administration

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 report details the development of Realistic Artificial Data (RAD) to improve highway safety modeling, specifically addressing the limitation that traditional models often fail to accurately capture underlying cause-and-effect relationships between roadway features and crashes. Sponsored by the Federal Highway Administration’s Exploratory Advanced Research Program, the project aimed to create synthetic datasets with known causal structures, allowing researchers to evaluate how well statistical and machine learning models reflect true safety dynamics. The study focused on two high-risk locations at diamond interchanges: ramp terminal left-turn crashes and speed change lane crashes. The researchers established a three-step framework to generate RAD. First, they identified contributing factors (roadway, traffic, and driver variables) using literature and observed data from Washington and Missouri. Second, they defined the effect of each factor on crash frequency based on published literature and the Highway Safety Manual. Third, they quantified the combined effects using a hierarchical Poisson approach to generate realistic crash counts and severity distributions. To facilitate access, a web-based software tool was developed, offering 196 pregenerated datasets and the ability to request custom datasets. Additionally, the team created virtual reality (VR) simulation testbeds by reconstructing 424 safety-critical events from the Strategic Highway Research Program 2 Naturalistic Driving Study, providing immersive visualizations for driver education and human factors research. The generated RAD datasets were used to test various modeling approaches. Three research teams developed models, including negative binomial regression and machine learning algorithms such as TabNet. A model evaluation rubric was applied, scoring models on descriptive analysis, selection, training, prediction accuracy, and inference. Model inference was weighted most heavily (50 of 100 points) to assess the ability to recover known causal relationships. Overall model scores ranged from 72 to 91, with performance differences primarily observed in the inference criteria. The VR testbeds were developed through a four-step process involving video analysis, crash diagramming, 3D modeling, and simulation reconstruction, offering aerial, 360-degree, and interactive test-drive views. The significance of this work lies in providing a rigorous testbed for validating safety models against known ground truths, thereby improving the reliability of crash modification factors and safety countermeasure evaluations. The generic RAD framework can be extended to other facilities, such as work zones or pedestrian areas. Furthermore, the VR testbeds support the USDOT’s Safe System Approach by enabling realistic driver education and the evaluation of behavioral interventions and vehicle safety features. This initiative enhances the toolkit available for data-driven safety analysis, moving beyond simple crash frequency estimation to a deeper understanding of crash causation.

Key finding

Realistic artificial data with predetermined causal relationships enable the evaluation of safety models' ability to accurately infer cause-and-effect relationships, which is often difficult with real-world data where ground truth is unknown.

Methodology

modeling

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.

• naturalistic crash near crash
• incidence prevalence
• induced exposure
• causation analyses
• telematics crash prediction
• crash typology

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: crash risk outcomes
• Methodological Resource: dataset resource
• Theoretical Contribution: computational model