I have a class-imbalanced dataset so I divided my data into positive and negative classes (10% pos-90%neg).
To model the data I planned to subset the negative data into 8 subsets and then create 8 databases by binding the positive data in to solve the class imbalance.
I ran 8 of the same model against the 8 subsetted databases and I was planning to use MuMin packages model.avg to average all the models however this doesn't work on "different data" i.e the subsets.
I have plotted the subsets and the data is representative in each database (i.e it is spatially distributed around the whole study area) SEE FIGURE.
I am wondering if I can use AIC to select the best model from the group of subsetted data as opposed to an ensemble method to account for the variance between the models and get a stacked prediction.
The model trends also differ slightly with the best model having the most precise prediction, with worse models being more generalised.
Is this acceptable?