I used deep learning for the answer I wanted to find.
I divided the results. I wanted to find into two. The result is 1 for success and 0 for failure.
So I used binary classification (binary_crossentropy)
If the proportion of 1 in the data is too low,
We can see the absurdity of the model processing all the answers to zero and boasting high accuracy. (accuracy: 0.9867, although it shows high accuracy, it was predicted to be zero even if data with a result of 1)
I want to find a deep learning method that can predict 1 even if the proportion of data in the correct answer is low
Does anyone know how to do it? Or give me some keywords so I can find them
Thank you for reading it.