I am working on machine learning model. The target is to learn the behavior of a black box which has a couple of inputs and one output signal. I have generated some training data by applying different input signals to black box however I am facing an issue, i.e., for a given set of input values the black box does not generate output signal of fixed value. Instead, the value of output signal varies a little. It is like adding a little variability to the output signal. My question is how shall I use my generated training data to model the black box behavior? Ideally, I would like to develop a NN model that for a given set of input signal values estimates the mean value of output signal as well as the min/max range.
I would start with some exploratory data analysis techniques before jumping into models. Visualizations of the data can save a lot of time.
I'll assume that features are fixed. Otherwise you might consider feature selection techniques.
Once you have explored the data, then you might start considering choices of model. You sort of answered your own question that you could train a neural network on the data generated from the black box. Just make sure you split the data into training, validation, and testing sets as usual for such ML projects.
But I would recommend starting with simpler models than typical neural networks. The failures of simpler models can hint at what improvements should be used, and you might luck out at a simpler model doing the trick.