Hi Lionel, You can look this paper : An Hybrid Method for Feature Selection Based on Multiobjective Optimization and Mutual Information

Hi,

Support vector machines can be used efficiently for variable selection in regression and in classification especially with linear data.

Recently I used the PCA to selection of Features on a set of 130 features and 114 thousand data objects.

when you run random forest or boosted tree algorithms for all features, they give you ordered list of features according to their importance.  These methods can calculate feature importance. PCA may be another option. Another one is the stepwise regression with either forward or backward selection.

genetic algorithm also can be applied