Machine learning

Machine learning refers to the methods by which a model infers patterns from examples, instead of applying rules written out explicitly. In fraud detection, its advantage is spotting combinations of signals that no rule would have anticipated, and adjusting when the methods used by fraudsters change. Its constraints are just as real: it needs labeled data in quantity, its performance degrades when reality moves away from the training data, which is model drift, and it reproduces the biases present in the history it learned from. Hence the requirement to be able to explain each decision.