Can merged datasets help in training money laundering detection models?