Performance Comparison of Real-Time Crash Prediction Models for Freeway Diverging Areas

Twelve interchange diverging areas from two multilane freeways in China were analyzed using image recognition to extract vehicle motion parameters. The dataset contains longitudinal and lateral conflict labels at 30-second intervals, used to develop and compare four machine learning models. It was created by Feng Tang and last updated on 2026-04-29.

Use Cases

• Training real-time crash prediction models based on vehicle motion parameters and surrogate safety measures.
• Analyzing spatiotemporal patterns of lateral versus longitudinal conflicts based on conflict labels at 30-second intervals.
• Interpreting key risk factor contributions for crash prediction using the SHAP framework mentioned in the description.
• Designing dangerous driving warning systems based on identified decisive factors like Modified Time-to-Collision.

Strengths

• Includes 48 vehicle motion parameters and surrogate safety measures extracted via image recognition.
• Applies the Extended Time-to-Collision metric, a validated two-dimensional metric for lateral conflicts.
• Compares four established machine learning model types: Random Forest, Neural Network, Support Vector Machine, and XGBoost.
• Results are interpretable via the SHAP framework, analyzing key risk factor contributions.

Limitations

• Row count is unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• Data is limited to 12 interchange diverging areas from two freeways in China, which may limit generalizability.

Provenance

Source

figshare

Collection Method

Data was gathered using image recognition technology on video from 12 interchange diverging areas.

Freshness

Last updated 2026-04-29 17:54:58; freshness should be verified.

Geography

Two multilane freeways in China.