Forecasting battles : New machine learning methods for predicting armed conflict
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Over the past decade, the field of conflict forecasting has undergone a remarkable metamorphosis, transforming from a series of isolated efforts with low predictive power into large, globe-spanning projects with impressive performance. However, despite this evolution, many challenges still remain. First, while we are good at predicting absolute risks, we are poor at predicting conflict dynamics (onsets, escalations, de-escalations and terminations). Second, we are over-reliant on spatio-temporal features and mechanistic models due to the nature of the event-data we use, thus excluding actor agency. Third, we do not handle either data or model uncertainty. Fourth, we are lagging behind the state-of-the-art in machine-learning. This dissertation attempts to resolve some of these salient difficulties, by contributing to six core elements of current-generation forecasting systems. First, time, by looking at the substantive effects and uncertainties of the temporal distance between data and forecast horizons. Second, space, by looking at the inherent uncertainties of high-resolution geospatial data and proposing a statistical method to address this. Third, feature space, by tackling the extreme feature sparsity in event-data and proposing a novel, deep active learning approach to mine features from existing large conflict-related text corpora. Fourth, substantive knowledge, by combining findings from the previous papers to take a fresh look at the microdynamics of conflict escalation. Fifth, the forecasting process itself, by building models that directly forecast from text, eliminating the intermediate step of manual data curation. Finally, the frontier of event-data, by looking at whether the news-media heavy way we collect violent fatal events can be extended to the collection of non-violent events. Methodologically, the dissertation introduces state-of-the art methods to the field, including the use of large language models, Gaussian processes, active learning and deep time series modelling. The six papers in the dissertation exhibit significant performance improvement, especially in forecasting dynamics.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.005 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.001 | 0.002 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.002 | 0.001 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it