LASSO feature selection result and selection of the best features

Question

Now i am applying Lasso for the purpose of feature selection and the result of features regression coefficients are mixed between (negative/positive/zero) values.

I know that "Any features which have non-zero regression coecients are “selected” by the LASSO algorithm".

Does it mean that i can use all positive and negative values and order it regarding the "value" only and neglecting if its positive or negative?

(Concern only about the magnitude regardless its direction)!

Please answer me and if u can recommend for me any simple document regarding that, attach it or set its name...

Answer 1

In LASSO regression, the L1 regularization can lead some coefficients to become 0. Only the coefficients that are non-zero are retained in the final model, regardless of their direction (i.e. the sign of the coefficients). The sign of coefficients tells you if the independent variable is positively or negatively related to the outcome.

So, to explicitly answer your question "can I use all variables with positive and negative coefficients in the final model", YES you can. Only the 0-coefficient variables are discarded from the model.

Have a look here and here for more info and example.