Machine learning and forecastsIndependent founder project
Global Risk Forecast
A LightGBM forecasting dashboard that turns historical incidents into regional risk views on a map.
Problem
It is hard to see where risk is higher from historical incident data.
Founder’s role
Dmitri trained a LightGBM model and built an interactive map of risk views.
Result
Regional risk is shown on a map, based on historical data rather than guesswork.
Problem
Sparse event logs are hard to read as a table. Teams need geographic, time-aware views and a model validated as a forecast, not a random-split notebook.
Solution
A pipeline from historical incidents to time-aware features, a LightGBM forecast, a FastAPI layer and an interactive map; time-split validation avoids leaking the future into training.
Result
Regional risk is explorable on a map, with timelines and simple explanations. These are statistical views of open data, not predictions of specific events.
