Please note that this a regression problem. We are going to predict the tax adjustment for each client.
If tax adjustement (target) = 0 -> it's not fraudulent
if tax adjustment (target) > 0 -> it's fraudulent