What Is Machine Learning and How Can it Help You?
February 25, 2016
Machine Learning is changing many industries including fraud risk management.
Machine learning is harnessing modern computing power to create an optimized statistical model. In traditional statistical modeling, a statistician needed to personally evaluate specific aspects of a dataset and interactions between variables to create a statistical model. With machine learning, the statisticians can leverage algorithms, programmed to run millions of calculations with brute force to try many options and suggest the best. The algorithm is programmed to learn from its calculations and use the technique coded within the algorithm to recommend a final equation for the model.
How has this impacted fraud risk management?
- It enables the new statistical models to be developed leveraging a significantly higher number of variables. The machine learning models that Accertify builds today for our Clients incorporate more than double the variables than prior logistic regression models used – meaning more information about a transaction can be used to make an overall better decision.
- It enables much more complex models to be built. Some recent models Accertify has created and implemented for Clients contained around 300 micro models within an overall model ensemble. This enables a broader coverage across a Client’s business line that historically was very difficult with prior model types.
- It speeds the time to create a model. Now, the majority of the time required to build a model is focused on careful understanding of a client’s data and processes, creating custom predictive characteristics and optimizing how the model is implemented to achieve the best results.
Machine Learning modeling has helped many Accertify clients reduce their review rate and improve their fraud capture rate – driving better business results.
About the Author
Kristin has over 17 years experience working in risk management and payments. At Accertify she leads a team responsible for improving risk decisioning.