Is keras suitable for such a task?

Question

Python

There is a data sample:

#1 219503 1726185 1531347661 3616 2 1 3 1 1.01 21 27 1.21 1.37 2.11 1.725 1.38 1.16 6 0 2 3 0 0 0 1 0 2 4 1 3 0 94 87 81 39 -1 -1 55 45 6 1 1
........................................
#N

It is necessary to predict the result on the basis of similar, but already new, received data.
30-40 parameters per input.
The output is 1 answer (probability from 0 to 1).
Example: I worked only with FANN, I would like the transition to be as simple as possible. Thanks in advance for the replies.
+ 0.99999117851257

Answer 1

In my opinion, the usual gradient boosting is more suitable for you. the same CatBoost or LightGBM, xgboost.
Neural networks for classification are inconvenient in that you need to edit the network architecture for good results, and this is a non-trivial task. They are suitable only when there is a ready-made, tested architecture for the task.