Logo Lanfrica

News-Signal-for-Food-Insecurity/news-signal-africa

Domaine:

socioeconomic

Type de record:

software
Créateur:
New
Hôte:
Replication code for: Does News Contain Meaningful Incremental Signal Beyond Autoregressive Structure? Evidence from Food Crisis Prediction in Sub-Saharan Africa # Does News Contain Meaningful Incremental Signal Beyond Autoregressive Structure? ## Evidence from Food Crisis Prediction in Sub-Saharan Africa This repository contains the complete replication code and pre-processed data for the paper. The pipeline predicts district-level IPC Phase ≥ 3 food crises 8 months ahead across Sub-Saharan African countries using GDELT news features evaluated against a pure autoregressive baseline under rolling temporal cross-validation. --- ## Repository Structure ``` news-signal-africa/ ├── data_preparation/ │ ├── build_dataset.py # Reconstructs DATA/dataset.parquet from raw sources │ ├── 01_prepare_ipc_reference.py # Prepares IPC reference data │ ├── 02a_stage2_aggregate_articles_monthly.py # GDELT article aggregation (requires raw GKG) │ ├── 03a_stage2_aggregate_locations_monthly.py # GDELT location aggregation (requires raw GKG) │ ├── 04a_stage2_create_ml_dataset.py # Merges aggregations → monthly parquet │ ├── download_gdelt_gkg_2020.py # Downloads raw GDELT GKG data │ ├── extract_gdelt_2020_locations.py # Extracts location mentions from GKG │ ├── fix_2020_gadm_names.py # Harmonises GADM district names │ ├── concat_locations.py # Concatenates location files │ └── extend_stage1_with_2020.py # Extends IPC dataset with 2020 data ├── DATA/ │ ├── dataset.parquet # Final modelling dataset (all scripts read from here) │ ├── dataset_summary.json # Dataset metadata │ ├── raw/ # Pre-processed source data │ ├── filtering/ # Coverage threshold analysis outputs │ ├── modelling/ # Fold-aware monthly news features │ └── shapefiles/gadm/ # GADM v4.1 ADM2 geometry (for Fig 1 maps) ├── 01_train_models.py # Primary 2-year rolling CV: AR-only vs AR+News ├── 02_rolli …