You may try these approach :
1.Use different configuration:
Divide your training set into two parts: one for training and the other for validation, then try to train and evaluate using different approaches and find what works the best.
2.A rule of thumb:
As per guesses, people can come up with a rough idea with the number of neurons in the hidden layer they are as follows:
Neurons should be placed between the input and output layers.
Should be set to something around (input+output)x(2/3).
It must not be larger than twice the size of the input layer.
3. An algorithm which dynamically adjusts the network configuration:
Using algorithms like cascade correlation which starts with a minimal network and eventually adds hidden nodes during the training, this makes your code simpler and enhances your performance.