Machine Learning – Buzz in the Industry


By Kristin Hoyne Gomes, Director, Decision Sciences - Accertify, Inc.

There is much buzz today about Machine Learning in the fraud/risk industry - and with that many new terms being thrown around to propose the best techniques and approaches to data modeling.  How can you makes sense of the hype to understand what is really important to focus on?  As in solving any challenge, there are often many approaches that may work and may even achieve very similar, good outcomes.

Machine Learning has brought many advantages to risk decisioning.  Accertify has found the greatest success in boosting predictive power leveraging ensemble models.  An ensemble model is one model composed of a combination of several smaller models.  This model type can easily incorporate risk decisioning on many different lines of business, geographies and other segments that each traditionally required an individual model.  Ensemble models can also generally leverage a larger number of variables than traditional models, which allows for the incorporation of more diverse risk factors and provides increased predictive power.  Due to its approach of multiple models combined into one overall model, ensemble models are also generally very robust to changing conditions and thereby maintain high predictive performance for longer than most monolithic models.

Important factors Accertify considered when selecting the algorithm and approach we use were:

  • Ability to build a robust model, incorporating many data elements to capture the risk differentiation available from a wide source of predictors
  • Ability to enable optimal predictions across the diverse channels, geographies and segments in which our customers operate
  • Ability to provide Clients clarity into the factors that drove the decision – displayed as Model Reason Codes on a transaction in the Accertify Interceptas® Data Management Platform
  • Speed of execution of the model transactionally – to ensure our Clients can leverage the power of the model in the milliseconds of processing before a decision must be supplied to their customers

Want to learn more?  Accertify will be launching a learning webinar on Modeling covering topics such as:

  • The differences between Machine Learning and traditional statistical models
  • How models are built and implemented and how they can improve decisioning

Look for more information in the coming month from your Account Manager.