What kind of NN I need to find ideal ranges and correlation between them?

Question

I’m new to NN and I’m trying to collect material and study. I’m getting through a general high level book, but I’m still struggling understanding what kind of NN I should go ‘deeper into’ for what is my idea.

I have some equipment that I can control from my computer. Each part of this equipment has a specific ID. Each ID accepts a value within a specific range, always the same for that specific ID: so for example ID1 might accept just 0 to 1, ID2 accepts 0-255, ID3 accepts from 0 to 3 etc. (the max number of ID/controllers is 255).

Now, the thing is that some ranges are better than others. So there’s a desirable outcome, that usually means some ranges work better than others for each ID/controller, and it also means that there are probably some correlations (for example the range 30 to 90 for ID1 usually works best when ID5 is in the range of 20 to 50).

The idea would be to have training data from random generation: so I generate a random value (within the full range of each ID) for the whole set of parameters, I decide if the outcome is good (0/1, good or bad), and then I generate once again random values for all the parameters, once again I label it if good or bad, etc. It all works with all parameters together, not one by one, so it’s like a snapshot for the status of the machine (all its IDs/parameters).

And to test the AI, I would have two choices: either let it generate ‘desired’ values directly, or have the AI after a quick random generator that let pass only what it considers as good.

My question is: what kind of NN would be recommended for this application? What would I need to study more? I’m proficient in Python so I would use that.

Considering the nature of the data (around 100 pairs of ID:value where value is within range that goes max from 0 to 255), I believe this should be an almost textbook example for using a NN. But maybe I’m wrong?

Any suggestion, link, direction would be much appreciated :-)

Answer 1

What you described seems like a pretty standard binary classification problem. There are many good algorithms, that are much simpler and more interpretable than NNs. I don't see why you would straight jump into NNs without first trying and fine-tuning e.g. Support Vector Machines (SVM), Random Forest(s), or some gradient boosting classifier like XGBoost. I suggest you have a look at the sklearn package in Python for the implementation of these.