If I have millions of data then it is usefull to apply resampling techniques like oversampling and downsampling in this case we dont have enough data of fraud class so my question is anyone here is using resampling techniques and getting good f-score?
Try it out, perhaps even try a combination of the two.
I think you should resample, especially since there are few fraud cases. Also as a general rule, when you have fewer events that are less likely to occur, you should either oversample the minority case or else undersample the majority case.
I support your suggestions