UK Road Traffic Accident Data for KSI Crash Prediction, 2020–2024

A dataset of United Kingdom traffic accident records from 2020 to 2024, reformulated for binary classification of killed or seriously injured (KSI) crashes versus slight injury crashes. The data was used to develop a LightGBM model with SMOTE for imbalance handling, safety-oriented threshold optimization, and probability calibration. The study was authored by Khanh Giang Le and last updated in April 2026.

Use Cases

• Training binary classification models for crash severity prediction based on the described KSI vs. slight injury outcome.
• Evaluating imbalance-handling techniques like SMOTE and class-weighted learning as referenced in the comparative analysis.
• Conducting model interpretability analysis using SHAP to identify key risk factors like speed limit and road class.
• Optimizing decision thresholds for safety-critical applications where recall of severe crashes is prioritized.
• Assessing probabilistic model calibration and reliability using metrics like the Brier score mentioned in the description.

Strengths

• Data covers a 5-year time range from 2020 to 2024.
• The associated study reports a 73% improvement in Recall(KSI) at an optimized threshold of 0.35.
• Model probability calibration achieved a Brier score of 0.190, indicating reliable risk estimation.
• Dataset is shared under a permissive CC-BY-4.0 license.

Limitations

• Row count is unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• The dataset is very small at 5.5 KB, indicating limited scope or a processed subset.

Provenance

Source

Nationwide United Kingdom traffic accident data.

Collection Method

Likely collected and aggregated from official UK traffic accident reports.

Time Range

2020–2024

Freshness

Last updated 2026-04-28 17:48:51; freshness should be verified.

Geography

United Kingdom