As per my knowledge, you can use Generalise Logistic Regression model and Linear Regression model which use T value and P value to find effective parameters.

As per my knowledge, you can use Generalise Logistic Regression model and Linear Regression model which use T value and P value to find effective parameters.

Hi Alireza Jafari ,

In theory, neural nets can able to ignore ineffective parameters by lowering the weights of connections between the layers. Simply, you can track these weights and bypass them to make your network less complex (assuming your inputs are normalized). But it might not be easy to apply.

Therefore it is better to select relevant features first, then feed your network with the selected features.

It is a good practice to view the most effective (linearly) features by calculating Pearson correlation with the target values.

For the selection of the minimum number of features, you can use LASSO or Genetic Algorithms by arranging the penalize and reward parameters strictly.