Multiple input single output supervised learning ANN problem

I have a dataset of 120 tuples giving a singular output. I want to implement ANN in estimating the input which is affecting the output most. A case of optimising the input to maximise the output.

ANN has in the input layer multiple features which can give n number of outputs. However in my case, the set of tuples gives a singular output. I want to know which combination of tuples is affecting the output most. Finding maxima or minima for the loss function.

The challenges here is there is no specific output mapped to a tuple. But a combination of 120 tuples give a singular output. In the proposed model i aim to feed further randomised combination of these tuples to train the model and later fetch the instruction(or tuples affecting the output most). Can we proceed by normalizing the output and then mapping the input tuples.

How can this be done?