It depends on your output layer and purpose of NN i.e. classification, regression. Very common implementations are as follow:

Binary Classification: One node, sigmoid activation.

Multiclass Classification: One node per class, softmax activation.

Multilabel Classification: One node per class, sigmoid activation.

for a more detailed answer, you can check out this link(https://machinelearningmastery.com/choose-an-activation-function-for-deep-learning/)

https://machinelearningmastery.com/choose-an-activation-function-for-deep-learning/#:~:text=The%20choice%20of%20activation%20function%20in%20the%20output%20layer%20will,deep%20learning%20neural%20network%20project.&text=The%20activation%20function%20for%20output,the%20type%20of%20prediction%20problem.

Is it possible for us to do a comparison study with at least three different datasets?

The best activation function will definitely be revealed.

please see this link :

https://arabicprogrammer.com/article/774691753/

Hello Chukwudi Paul Obite ,

I want to add two pieces of information that might be worked for your problem.

All activation functions are non-linear and for modern deep learning, ReLU is extreme popular because of it's simplicity.