Data Science Series: Explainable AI - Bringing NeuroDecision to the UK

Cracking open the black box means giving credit applicants the key factors that underpin credit decisions and ethically expanding credit access for all consumers.



The Equifax Data Science Lab in Atlanta, US, invented NeuroDecision® Technology – the world’s first explainable AI patent in credit risk – in order to help businesses approve more customers while keeping risk levels constant, or decrease delinquency rates while holding  approval rates constant. Now this innovative technology for predicting behaviour is available to our UK customers.


Even though it’s a key part of many businesses, reliably predicting the future behavior of specific customers is not easy. Today we have more consumer data than ever to draw on, but of course, data alone is not enough, we need a means of turning this data into actionable insights. 


For many lenders, Logistic Regression remains the tool of choice in credit risk modelling; it’s a proven and explainable modelling technique. Using logistic regression we can predict if a credit applicant will default on a loan. However, logistic regression has its limitations, it shows us a linear relationship between attributes and outcome in our predictive model. But what about those credit applicants who don’t fit so neatly, where the relationship between the attributes we’ve chosen doesn’t fit the linear decision boundary as described by our logistic regression model? 


Fortunately, the capability to better capture these interaction effects exists within the neural network technique, a machine learning algorithm that is powerful enough to capture finer distinctions and make better predictions, resulting in strong uplift for some applications. So why then are neural networks not common practice in credit risk modelling? Neural networks are often called black box models because they are not explainable, this is a real challenge in regulated business environments. 


What if we could harness the power of neural networks for regulated business applications? Remarkably, this is exactly what researchers in the Equifax Data and Analytics Lab have done. The result, NeuroDecision, represents a deep and careful reworking of the neural networking modelling approach. NDT combines the cutting edge performance of neural networks with the explainability required to adhere to industry best practices and comply with complex regulations. 


Matthew Turner is from Equifax’s Data Science Lab and co-inventor of NeuroDecision. In comparison to other alternate explainable AI offerings in the market, he says that “When making a decision impacting a consumer’s financial well-being, a data scientist always needs to explain and justify the reasons underpinning the decisions... Only NDT can truly capture what the model is doing - not explanations that hold in general or explanations that work on average, where individual results may vary. Instead the explanations that we give are personalised to an individual consumer. And if that individual consumer responds favorably to the explanations, their score will improve. And if they respond negatively to explanations, their score will decrease.” 


Watch the video (an intro to NeuroDecision) >



Contact us >