Omar Mohamed, Faiza (2026) Predicting Political Violence in West Africa: Does GDELT Improve Short-Term Forecasts Beyond ACLED? Bachelor thesis, Data Science and Society (DSS).
|
PDF
BA5607132FOmarMohamed.pdf Download (1MB) | Preview |
Abstract
This thesis examines whether international media attention variables derived from GDELT improve short-term forecasts of political violence in West Africa beyond what conflict history alone can predict. Using a country-week panel covering Mali, Burkina Faso, Niger, Nigeria, and Ghana from 2015 to 2023, two nested negative binomial regression models are compared across one-, two-, and four-week forecast horizons. The baseline model uses lagged ACLED conflict-history variables, while the extended model adds lagged GDELT measures of media tone, Goldstein scores, and reporting volume. The results show that the extended model consistently outperforms the baseline across all forecast horizons. However, the improvement is uneven across countries. Nigeria accounts for most of the gain in the pooled model, while the robustness check excluding Nigeria shows stronger improvements for Mali and Burkina Faso. This suggests that media-derived indicators are useful for conflict forecasting, but that their predictive value depends strongly on the intensity and visibility of the conflict environment. The thesis concludes that GDELT can complement ACLED in short-term forecasting, although pooled models should be interpreted carefully when countries differ substantially in conflict dynamics and media exposure.
| Item Type: | Thesis (Bachelor) |
|---|---|
| Name supervisor: | Haleem, N. |
| Date Deposited: | 10 Jun 2026 10:57 |
| Last Modified: | 10 Jun 2026 10:57 |
| URI: | https://campus-fryslan.studenttheses.ub.rug.nl/id/eprint/799 |
Actions (login required)
 |
View Item |