Features Analytics Services

 

 

eyeDES® one model does not fit all

The Model Development Process

The modeling process through six stages to develop effective, custom predictive models.

 

The six stages are:

 

  • Data gathering and pre-processing
  • Data analysis
  • Data enrichment: Feature design and calculation
  • Data segmentation
  • Machine learning model development per data segment
  • Model optimization and delivery

eyeDES® - Custom Models that evolve with the data

Herein lies one of the key aspects of Features Analytics and the eyeDES® streamlined machine learning technology: the focus on enriching the customer data with sophisticated features able to detect hidden patterns but also evolve with the data by adapting in real-time. This removes bias toward the requirement for certain data elements being needed for the models to perform effectively, as all types of data features are analyzed historically in their own context. The fraud and non-fraud patterns that the features are describing are used in combination with machine learning algorithms to develop predictive models - that allow to automatically assess decisions and outcomes regarding the likelihood of fraud.

 

This means that eyeDES® predictive models can be applied and utilized by different types of organizations, from eCommerce merchants to acquirers, processors and issuing banks. This focus on custom modeling utilizes the data variables an organization has available uncovering patterns and risk signals specific to the organization. Although no single variable or feature can fully discriminate between fraud and non-fraud records, by combining several of them and applying state of the art machine learning algorithms, complex behavior patterns and trends can be detected and followed by the evolving models.

 

Multiple Models – one model cannot fit all your needs

Our team of data scientists develops multiple models to different segments or channels of your business. This could include separate models for different countries or regions, or different levels of risk based on factors such as customer age or purchase amount. These models can be applied in addition to or in place of analytic models your organization may have in place, each providing a risk score and reasons contributing to the transaction assessment. The level of performance can be maintained or improved for each model segment, in contrast to fixed models that utilize all transactions indiscriminately. The right combination of the models will deliver an overall optimum performance of your solution.

 

Different machine learning algorithms may be required to get the most performance out of each individual data segment. The Features Analytics streamlined machine learning infrastructure allows us to rapidly implement, test and decide on the multitude of possible variable, data segment and machine learning algorithm combinations in order to deliver the maximum performance within a limited amount of time.