What is the difference between machine learning and rules-based detection?
A good fraud strategy uses a combination of machine learning for identifying indicators of fraud, and rules that enable alerts to be triggered on unusual patterns of behaviour that may include these indicators.<br/><br/>
Machine learning works best when there are large volumes of data available to train the fraud models. The models need a combination of 'true negatives' (data where there is no fraud) and 'true positives' (data where there is fraud). From this, a series of algorithms can be developed that are able to alert suspicious behaviour.<br/><br/>
Rule-based detection is helpful for enabling simplistic controls and some organisations are able to prevent over 70% of attempted fraud just using simplistic rules around volume, velocity, location and device profiling.