Crash Severity Prediction for Emergency Dispatch

This dataset contains the final output artifacts and diagnostic figures generated during the "Crash Severity Prediction for Emergency Dispatch" experiment. Developed for the Data Stewardship 2026 (Part 3) course, this project models road collision severity (categorized as Slight, Serious, or Fatal) using a Random Forest classifier. The models were trained and evaluated on the 2023 UK Road Safety Open Data (STATS19).

These outputs provide full transparency into the model's performance and the exploratory data analysis (EDA) conducted prior to training, ensuring the reproducibility and evaluability of the machine learning pipeline.

Dataset Contents

The deposit is organized into two main categories:

1. Evaluation Reports (reports/) Tabular data detailing the model's predictive performance:

• report-classification-metrics-2023-v1.csv: Comprehensive evaluation metrics (Precision, Recall, F1-Score) for each severity class.

• report-predictions-test-2023-v1.csv: The raw model predictions mapped against the actual test set values.

2. Visual Diagnostics & EDA (figures/) Graphical representations of the data distribution and model behavior:

• Model Diagnostics: A confusion matrix (fig-confusion-matrix-2023-v1.png) detailing prediction accuracies and misclassifications, alongside a feature importance chart (fig-feature-importance-2023-v1.png).

• Target Distribution: A visualization of the class imbalance in the dataset (fig-severity-distribution-2023-v1.png).

• Feature Histograms: Distribution plots for all scene conditions used as features in the model, including day of the week, light conditions, number of vehicles, road surface conditions, road type, speed limit, vehicle type, and weather conditions.

Methodology

The data was processed and modeled using Python (scikit-learn, pandas). The original raw data was flattened from three relational tables (collision, casualty, vehicle) into a single feature matrix before being split into training, validation, and test sets.