How to organize monitoring and support of web applications based on neural networks?

Question

Machine learning

Good afternoon community.
As part of the study, I asked myself how to organize monitoring and support for web applications based, for example, on neural networks.
I'll give you an example. There is a web application that works in real time. And there are many parameters that can be changed (such as the pool of connections to the database for reading, the pool of connections for writing, HTTP connections, etc.) to achieve maximum efficiency (say, the number of successful requests per unit of time).
Tell me how to properly organize work with time series and how what are the output signs (outputs)?
UPD0.Commentators do not see the phrase "within the study." I would like to emphasize this again. I'm not going to shove it into the prod, sell, conduct seminars / trainings, start a separate twitter account. I'm interested in testing the theory.
UPD1. Yes, I'm not experienced in the National Assembly. An online course, a couple of kaggle puzzles and a small self-written framework (written for fun, to figure out how it works from the inside.)

Answer 1

First you need to have such an application in which there are a lot of parameters that you can change and a lot of metrics that you will use to train.
It seems to me unlikely that there is an application suitable for training a neural network.
Well, that is, there are such applications on Google or Facebook.
There are also cloud providers.
But where do you get such things in ordinary life in order to train a neural network?
As soon as you find it, all other questions will disappear by themselves.

Answer 2

Change the input parameters (number of connections, etc.) and fix the output: time and number of errors - both parameters should tend to zero.
Here, even NN is not needed: just create tests and write down the output values ​​for the input and choose the optimal ones.