oluwafemidiakhoa/lassaai

AI-powered Lassa fever outbreak prediction and clinical decision support for Nigeria # LassaAI **An open, honestly-benchmarked national early-warning model and clinical decision-support aid for Lassa fever in Nigeria.** Built by Oluwafemi Idiakhoa · Houston, TX Open source · MIT License · 2026 --- ## What LassaAI is (and what it isn't) Lassa fever is endemic in Nigeria and the national response is largely reactive. LassaAI is an **open, reproducible pipeline** that asks a modest, honest question: can we anticipate weeks of *elevated* national transmission a month ahead — and does a model actually beat a trivial baseline? **It is:** a national weekly early-warning proof of concept on **real, cross-validated NCDC data**, benchmarked against naive baselines, plus a point-of-care clinical-probability aid and an antiviral-evidence tracker. **It is not:** a validated clinical or operational forecasting system. The model beats a naive persistence baseline only slightly; genuine skill can only be shown prospectively. We make **no claim of exceptional accuracy.** > **Preprint (open access):** doi.org. Manuscript source: `docs/lassa/paper-draft.md`. --- ## Live platform | Feature | URL | |---|---| | Outbreak dashboard | gailabai.com | | Clinical copilot | gailabai.com | | One-page summary | gailabai.com | --- ## National early-warning model ### Data (real, triple-cross-validated) - **National weekly incidence, 2020–2025** — NCDC Weekly Epidemiological Reports, via a provenance-tracked public compilation *and* our own independent re-extraction (`scripts/lassa/rebuild-ncdc-real.py`). The two extractions and NCDC's published annual totals agree within **~1%** (e.g. 2020: 1,190 vs 1,189; 2024: 1,311 vs 1,309). - **Annual confirmed cases:** 2020 = 1,190 · 2021 = 511 · 2022 = 1,042 · 2023 = 1,271 · 2024 = 1,311 · 2025 = 1,131 (total 6,456). - **Individual-level cross-check (2018–2021):** the de-identified SORMAS dataset (20,062 records, CC-BY-4.0) — its …