Classification problem (X, Y, Z) input accelerometer and (0 or 1) target

Hello ,

I would like to build a neuron network (multilayer perceptron) using the tools of Matlab.

Personally, I used the tool nntool feed forward which I used two layers; the first layer with sigmoid activation function and the second sigmoid layer

But I do not know if it is well adapted to my data even with a lot of tests.

According to the regression representation the data do not fall on the 45 degree axis meaning that we have a good training.

Could you please tell me what is the most appropriate neural network in your opinion for my data? where the inputs (X, Y, Z) are accelerometer coordinates normalize between (0/1), The target are (acceleration, braking, left turn, right, vibration, ...)

For my part, I coded it in 0 or 1 (example (1,0,1,0,0) which means that I accelerated and I turned left).

  Here is the classification used? that it network use and activation function. learnGDM / learnGD

MSE / SSE

 

Thanking you for your answer,

Regards,

Lylia BELLABIOD