For some time now, machine learning, artificial intelligence, and other futuristic-like technologies have been a reserve for leading tech companies including Facebook, Google, Microsoft, Apple, and Amazon. Nevertheless, the situation appears to be changing as traditional companies like SAS and Equifax seem to be joining the game.
Recently Equifax unveiled a machine-learning credit scoring system designed to utilize network modeling to assist clients in assessing risk more accurately. The NeuroDecison Technology solves a previous impediment faced by the company through neural network applications. The technology will help both regulators and customers understand better commercial and consumer risk models.
Peter Maynard, Equifax’s senior vice president of global analytics, said in an interview that Equifax realized some years back that it was not receiving the adequate statistical lift from its traditional/ previous credit scoring methods. Subsequently, the company began to embrace the use of deep-learning technology.
The report found out that modern machine-learning technologies that have more accurate results like deep neural networks were seen to be uninterpretable. This scenario posed an obstacle for any company that wanted to utilize them. Furthermore, the complexity added more problems for Equifax.
Faced with this challenge, Maynard’s team at Equifax found a way to make the neural networks interpretable. They created a mathematical proof to show Equifax can come up with a neural network solution that can be entirely interpretable for regulatory reasons. Every input can map into the neural network’s hidden layer. Aside from imposing a set of criteria that allows the company deduce the qualities coming into the final model, the team stripped the black box apart in a bid to get interpretable results.
In the report, Maynard acknowledged that the neural net has enhanced its capacity to create predictive models by 15 percent. The more sophisticated the data, the better the advancement. Identifying the best segment can sometimes take 20% of the time taken in building a model. The neural network does this task for you.
Equifax is not the only credit bureau utilizing AI solutions. In fact, FICO uses artificial intelligence (AI) to first group customers into more identical categories of credit risk models before generating more accurate scorecards with a better interpretation of outcomes. On the other hand, VantageScore utilizes ML algorithms in assessing client risk and assigning scores, more precisely to credit invisible clients who lack recently updated credit cards.
As Equifax progresses with its recovery journey after a data breach that cost it more than $114, it acknowledges the importance of reestablishing trust in its commodities and staying abreast of the developments in fintech.
According to Peter Maynard, being regulatory compliant makes Equifax’s machine-learning application unique. The company’s framework produces reason codes with outcomes that boost model accuracy. Despite other companies using machine learning methods in segments of the model building process, the last score is not related to a machine – learning regulatory compliant system/model. These methodologies lack the transparency and explain-ability that is needed for both regulatory and compliance purposes.
The research for ML and AI applications intended for credit scoring system began in 2015. The company looks forward to introducing additional modeling technologies to the market.