Optimizers -- Nadam/Adam learning

Hi everyone,

I’m currently working a kaggle competition using structured data with very skewed data.
As a start point, I tried to run a simple MLP using a Nadam optimizers. The issue is that I could not lift up the ground. My model kept giving me the majority class as prediction and a AUC of 0.5.

I changed the optimizer to Adam and my AUC rose and showed some good results.

What could be the reason for this difference of results ?