This repository contains the public-release code, data snapshots, and model artifacts supporting the Final Deliverable Slides.
The project explores an automated election-risk assessment workflow for the Red Cross SARGE context. It combines country/election features, public API signals, article retrieval, and a SARGE-aligned modeling strategy that predicts likelihood and impact before mapping the combined score into risk categories.
articles.py- WorldNewsAPI-based election article retrieval with translated search keywords.modeling/country_features.py- country-level governance, economic, press freedom, regime, and historical-violence feature builder.modeling/download_acled_events.py- ACLED event downloader for violence history features.modeling/source_inventory.py- source inventory and public API feature extraction utilities.modeling/clean_training_features.py- cleans and validates model inputs.modeling/train_sarge_subscore_models.py- shared likelihood and impact modeling helpers used by the combined pipeline.modeling/train_sarge_combined_pipeline.py- trains the final combined quantitative risk pipeline described in the final deliverable.data/- cleaned release data, model reports, predictions, feature importance tables, and source inventory outputs.models/- fitted SARGE combined-pipeline model artifacts.
Use Python 3.11+.
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt