Why do I get good accuracy with IRIS dataset with a single hidden node?

Question

I have a minimal example of a neural network with a back-propagation trainer, testing it on the IRIS data set. I started off with 7 hidden nodes and it worked well.

I lowered the number of nodes in the hidden layer to 1 (expecting it to fail), but was surprised to see that the accuracy went up.

I set up the experiment in azure ml, just to validate that it wasn't my code. Same thing there, 98.3333% accuracy with a single hidden node.

Can anyone explain to me what is happening here?

Answer 1

Iris is very predictable and also that there are fewer features in the data set. Also, there are high linear correlations. These facts point to a less complex, linear function which yields good results. Since you have used, you are nearly using a linear model.