Forecasting / Dashboards · Independent project
Global Risk Forecast: Regional Incident Forecasting Dashboard
Dmitri Derevjanko built the Global Risk Forecast dashboard, a LightGBM forecasting product that turns historical incident data into regional risk views on a map.
This is one of 10+ independent public projects by Dmitri Derevjanko. 100+ additional workflows were delivered professionally, mostly under confidentiality, and are not documented here.
Open live demo View GitHub Back to portfolio
Problem
Sparse event logs are hard to read as a table. Teams need geographic, time-aware views and a model that is validated as a forecast, not a random-split classification notebook.
Solution
A pipeline from GTD-style incident history to time-aware features, LightGBM forecasts, a FastAPI layer, and an interactive map with scenario-style exploration and lightweight explanations.
What Dmitri Derevjanko built
Dmitri engineered leakage-aware features, trained and validated LightGBM with time splits, built the serving API, the map UI, and published the code in the ai-risk-clean repository.
Architecture
- Historical incidents become a supervised dataset with lags, rolling windows, and seasonality proxies.
- Time-split validation avoids leaking the future into training.
- FastAPI serves region and horizon queries plus historical context for the UI.
- The map shows risk buckets, drill-down timelines, and simple “what changed” descriptors rather than raw SHAP dumps.
Technologies
- Python
- LightGBM
- FastAPI
- Pandas
- Interactive maps
- JavaScript
Result
An interactive forecast dashboard and public GitHub repo that make regional risk explorable without living in a notebook.
Limitations
- Forecasts are statistical views of historical open data, not predictions of specific future events and not an intelligence product.
- Sparse regions remain uncertain; low-data areas can look stable simply because little was recorded.
- The dashboard is for exploration and monitoring style workflows, not operational alerting.
Live demo and source
Live demo: https://ai-risk.dmitriderevjanko.com
GitHub: https://github.com/DmitriDerevjanko/ai-risk-clean
Portfolio case study: https://dmitriderevjanko.com/projects/global-risk-forecast/