Getting 88/12 percent train/test ratio with rebalancing


I’ve collected 80 accelerometer samples in four groups (10s each sample). When I rebalance for testing, I get an 88/12 percent split. Three of the groups supply 2 samples and the other group supplies 4 samples. I’m trying to figure out how that might happen? I was expecting four samples from each group in the test set for 20%.

Project ID:


Context/Use case:

Hello @robertgallup,

The rebalancing splits your data around 80/20.
It’s an approximation and it can vary based on your sample length.

Feel free to manually move some of your samples from your training set to your test set.



Thanks for the reply! Moving the samples manually was an easy way to get the balance I wanted.


1 Like