if you attach the reference, maybe you'll get some valuable advice, but "k-greedy for select feature on classification" does not really ring a bell to me
(well very slightly, just keeping the k features with the strongest correlation (loosely speaking) with the target ?)
anyway : please attach the reference
.
one last word ; if you do not know where to find it (whatever "it" is), then look in WEKA ... it might well be sitting there, waiting for you