artificial intelligence - How to eliminate "unnecessary" values to a neural network? -

my professor asked class make neural network try predict if breast cancer benign or malignant. i'm using breast cancer wisconsin (diagnostic) data set.

as tip doing professor said not 30 atributes needs used input (there 32, first 2 id , diagnosis), want ask is: how supposed take 30 inputs (that create 100+ weights depending on how many neurons use) , them lesser number?

i've found how "prune" neural net, don't think that's want. i'm not trying eliminate unnecessary neurons, shrink input itself.

ps: sorry english errors, it's not native language.

that question being under research right now. called feature selection , there techniques already. 1 principal componetns analysis (pca) reduces dimensionality of dataset taking feature keeps variance. thing can see if there highly corelated variables. if 2 inputs highly correlated may mean carry same information may remove without worsen performance of classifier. third technique use deep-learning technique tries learn features later used feed trainer. more info deep learning , pca can found here http://deeplearning.stanford.edu/wiki/index.php/main_page