Functional Needs Repair Vs Non Functional -> Resample due to Class Imbalance?


The Functional Needs Repair is less than 5000 responses compared to Functional and Non-Functional categories.
I believe it is a classic case of class imbalance (approximately 8 and 6 times) I think.

Is non-functional and functional needs repair that significantly different? In both cases, repair is still needed?
Is it wise for me to resample it and have functional needs repair under non-functional and train the model that way?

Any and all advices are welcome. I am quite a newbie here!

If whatever model/estimator supports weights, you can easily “solve” the imbalance by attributing weights to each sample inversely proportional to their class frequencies.

But not always such tricks are useful. If the testing dataset is also imbalance then you may suffer by changing the priors of the training set. In general, when using metrics such as accuracy, I would think you want to make sure you have the same priors in the training that you’ll have in the testing set.