See the code that builds and evaluates multilayer perceptron and that comes with the textbook at [link]. Build
3 versions of the model with the following activation functions: ReLU, LeakyReLU, Sigmoid.
a) The model built with ReLU activation function will likely have issues with vanishing gradients.
b) The model built with LeakyReLU activation function is prone to overfitting.
c) The model built with Sigmoid activation function is suitable for binary classification problems.
d) All three versions of the model will have similar performance regardless of the activation function used.