I have a xray dataset containing 14 classes. The dataset is skewed.
But the predictions are highly biased towards the majority class. see int the image
The number above is the f1 score which is very low.
How can I approach this?
I am using resnet18. More complex architectures are overfitting much more