Predicting PM-related mortality with insufficient training data

AI Discussions
, ,
1

Hi, all!

My current challenge is the following: we want to predict future mortality from PM2.5/PM10 and population data. We have 20 years of population data and PM2.5/PM10 on a grid, but only 1 number (the mortality) per year. My idea was to treat the information as a picture, but instead of RGB use PM2.5, PM10 and population. And rather than telling me that the picture is a cat, a dog or a parrot, I want my programme to give me a number.

This gives us only 20 training images, which is probably too small. I thought that maybe I could rotate the images and flip them, giving me an extra 100 training images. So I now have two questions:

  1. How would I rotate or flip my images?
  2. Will 120 images be enough to train my model?

Cheers,
Dave

1 Like