How to compare the results of my traffic prediction?

I have an LSTM model with three different activation functions and three different optimizers. I obtained the trained and tested results for each of them using MAE and MAPE metrics. Now, I want to compare them to see which one works better. Should I just consider the tested values and compare them, or do I have to calculate the difference between the trained and tested values and then compare the result of this difference?

I considered only the tested values for each model.

Yes, comparing the tested values for MAE and MAPE across your models is the appropriate approach. This direct comparison allows you to identify which model performs best on unseen data. There’s no need to calculate the difference between trained and tested values for this comparison. Focus on the model with the lowest MAE and MAPE on the test set to determine the best performer for your traffic prediction task.