Hidden layer and hidden neuron

I have 106587 data (11843rows *9 columns). I need to predict one parameter (output) using 8 other parameters (inputs).

1) How many hidden layers should I use (2 or 3) ?

2) How many hidden neurons should I assign for each hidden layer?

I tried 2 hidden layers with random (1-20) hidden neurons each layer but huge amount of MSE obtained for training, validation and testing. Please explain why this happens?

Answer 1

[Burhan Burak AKMAN]

Increase amount of hidden neurons to 100-200 and re-train. May be it will help.

[Pavithra Chinatamby]

I tried this method, but it took more than two hours per run and the MSE is almost the same when I use 1-20 hidden neurons.