Have you taken this into account: When many inputs are available, the choice of network architecture is especially important. For example, MLPs tend to be better at ignoring irrelevant inputs than are some RBF networks. Having many inputs also reduces the number of hidden units that you can use, since the number of weights connecting an input layer and a hidden layer is equal to the product of the number of units in each. For example, if you have five inputs, using 100 hidden units might be quite practical. But if you have 100 inputs and try to use 100 hidden units, you will have more than 10,000 weights in the network and training will take a very long time. source: https://documentation.sas.com/doc/en/emref/15.2/p0zbgj1tu3h1uhn1x6regixbdg7v.htm You may also want to look at: https://www.sas.com/content/dam/SAS/support/en/sas-global-forum-proceedings/2019/3274-2019.pdf
... View more