Crisis and resilience · Parked
CATIA, catastrophe risk against money
Hurricane, flood and wildfire risk scored against financial impact, joining actuarial method to machine learning.
Python ·
What it found
The modelling outran the data that could legally be obtained. Worth knowing before anybody starts this again: the blocker is procurement and privacy, not method.
The work
What was actually built
Catastrophe modelling is a mature actuarial discipline with its own literature, and the interesting question was whether learned models add anything to it or merely restate it with more variance.
CATIA got far enough to answer that the approach is sound and the data is the obstacle. The hazard data that would make it work is either commercial, restricted, or granular enough to be personally identifying.
Parked rather than abandoned. The code is public and the reason it stopped is recorded, which is more useful than a half-finished model.
See it
Open it, or read the code.
github.com. Opens in a new tab on an external site.