On the strong consistency of approximated M-estimators