Abstract:

Using US generated synthetic data set, this study explores the potential of three hybrid risk-based intelligent algorithms for onset prediction of financial crimes. The algorithms were selected in consideration based on the theory of imprecision, data uncertainties, knowledge ambiguity, and changes of fraudster strategy and tactical moves of fund. As transactions of data are highly imbalanced, data augmentation techniques have been introduced prior to training and testing phases of selected algorithms’ evaluation of their simulated performance. Performance metrics (suspicious score, and ML performance measures) statistically indicated that the three selected algorithms have yielded some predictability of patterns of potential AML activities, though imperfect.