Hello,
I have done the homogeneity test of the precipitation data-set. It is showing non homogeneity. I am going to use the precipitation data-set in Artificial Neural Network. Is it needed to transform the data-set in to homogeneous data-set before applying it to ANN?
Thanks & Regards,
Juthika
Hi Juthika! Something usual to do before working with ANNs is to scale the inputs to have mean 0 and a variance 1. Linear decorrelation/whitening/PCA could help as well. Just in case, correctly encoding the categorical input data is another preprocessing task to take into account.
Hi Juthika Roy.
A stardard procedure to use an artificial neural network is to normalize to interval [0, 1] or [-1, 1], depending on the type of activation function used. By the way, don't forget to rescale the output data.
if I understand the question correctly, you want to know whether the distribution of inputs should be balanced, right?
In principle, it is recommended to have a balanced input so that the number of A cases is comparable to a number of B cases. The logic behind this is simple - if the distribution is strongly skewed, the network will be well trained in recognizing the most frequent type, but will be less certain with the second one, and the number of false solutions might be high.
- Badges
- Science topic