I interpret your question in a way that you would like to improve your result. Being definitely not experienced in machine learning but simply looking at you data, I find:

From this you are using the first 70 tuples for learning and the last 30 for testing. I guess one can not expect under these conditions a good result.

As a probably better approach you might try:

train = RandomSample[tuples, 70];
test = Complement[tuples, train];
Length /@ {train, test}
(*  Out:  {70,30}  *)

Using this I get (e.g.):

Correlation[YPrediction, YObserved]
(*  Out:  0.9251388656090284`  *)

Daniel,

I tried different ways, I could not get the expected a fixed output. Could you please show me how I can do it with RandomSeeding option for the above example?

Thanks a lot.