Logo Lanfrica

ParisaGhahreman/Road-Accident-Severity-XAI

Domaine:

peace and security

Type de record:

project
Créateur:
Par
Hôte:
Cross-country road accident severity prediction and explainable AI framework using UK, France, and Ethiopia crash data # Explainable Machine Learning for Road Accident Severity Prediction Cross-country road-accident severity modeling and explainable AI framework using harmonized crash data from the **UK, France, and Ethiopia**. The workflow covers data audit, harmonization, cost-sensitive multiclass modeling, external/cross-country validation, domain-shift analysis, SHAP interpretation, driver-factor analysis, and an XAI-guided intervention-priority framework. ## Repository structure - `notebooks/01_Data_Audit.ipynb` — source-data audit and target/feature compatibility checks. - `notebooks/02_Data_Preparation.ipynb` — three-country harmonization and processed-data generation. - `notebooks/03_Modeling_and_XAI.ipynb` — baseline UK model comparison. - `notebooks/04_External_Validation_France.ipynb` — cost-sensitive model selection and UK→France external validation. - `notebooks/05_Ethiopia_Driver_Factors.ipynb` — driver-factor ablation and repeated CV in Ethiopia. - `notebooks/06_Cross_Country_Comparative_Analysis.ipynb` — three-country domain shift and six-direction transfer. - `notebooks/07_SHAP_Explainability.ipynb` — UK/Ethiopia SHAP interpretation and stability analysis. - `notebooks/08_Intervention_Framework.ipynb` — evidence-to-action prioritization, transferability, and sensitivity analysis. - `outputs/article/` — curated manuscript-ready figures and tables. - `outputs/tables/` and `outputs/figures/` — full reproducibility outputs. ## Data Raw and processed datasets are intentionally excluded from this repository. The notebooks expect source files under `data/raw/` and generate harmonized files under `data/processed/`. See `data/README.md` for the expected local layout and `DATA_SOURCES.md` for the exact download pages, citations, and reuse terms. Datasets used in the analysis: - UK STATS19 road-safety data - France BAAC road-accident data - Ethiopia road-accident dataset used in the project ## Main analytical design The target is harmonized to three severity classes: * …