How to test a neural network\filter training data?

Question

Neural networks

Good day.
Graduate work. There is a task of recognizing a disease according to statistical data.
There is a sample for three groups (healthy people, sick people, sick people with complications), a small one (from a few dozen for sick people and a hundred+ for healthy people).
Tried to solve the problem using brain.js. After training, the network gives completely inadequate estimates. Most often, it produces the same group at the output, regardless of the example given as input. A large number of iterations and an increase in the number of layers gives little progress. The minimum error achieved in the learning process is 0.1
The question is how to check the adequacy of the network (maybe I'm not teaching correctly), maybe there is test data that can be adapted to my case? Either the adequacy of the data (maybe processed incorrectly, or you need to throw out some parameters). There are only 45 parameters, I normalize before serving, divide by the maximum so that they fall into the range [-1; 1], I serve everything in one array - 150 people, 50 from each group.
In the most extreme case, you need another, not difficult to master, method or tool that can solve the problem.
Well, in the best traditions - you need it yesterday :(

Answer 1

Before taking on such work, and even with incomprehensible tools, you need to have knowledge of the design of the National Assembly. Each task is specific and can be solved in different ways when designing a neural network.
Creating a neural network is the development of a model through "pure" mathematics and formulas 95% of the time. And only 5% - algorithm coding.
Without fundamental knowledge of the National Assembly - the result is predictable.
Do you need yesterday? -> freelancing.