Time series forecasting

I use the same models in Time series forecasting  |  TensorFlow Core with different data and the strange things the baseline and linear give the best performance. and I notice when adding layers or units to the models the performance worsens. can anyone explain why?

Thank you so much for your attention and participation.

I’m afraid there can be tens of reasons why this does work not as well as expected.
Did you use the exact same code and just plugged a different dataset to it?
How similar/different are you data/dataset from the ones in the TF documentation?

Everything is the same except the dataset, and i use different normalization methods, that improved my results. I use max absolute values for normalization.
Everything else is the same also i focus on single step with on label column