Explainable AI in Fraud Detection - A Back to the Future Story
Explainable AI has become a hot topic in light of the European Union’s GDPR and society’s increasing reliance on AI systems. How does one of FICO's original Explainable AI innovati…
In 1992, Falcon Fraud Manager was introduced: a neural network-based fraud detection system that detects fraudulent payment card transactions in real-time. Since then, despite the proliferation of fraud types, there has been a dramatic reduction in payment card fraud losses across the globe.
One of the reasons we developed scoring technology was to help analysts take action on the transactions most likely to be fraudulent. We introduced the Reason Reporter, which, as the name suggests, provide reasons associated with the neural network scores Falcon produces. This capability is, in fact, Explainable Artificial Intelligence (XAI), a topic that recently has become a hot one in light of the European Union’s General Data Protection Regulation (GDPR) and society’s increasing reliance on AI systems.
In 1996 we filed a patent for Reason Reporter—indicative of how long, in fact, FICO has been working with Explainable AI. Today, it's still performing very well, and in fact outperforming some newer techniques in the Explainable AI realm.
How XAI works in Reason Reporter
In the context of Falcon, Reason Reporter is designed to provide explanations for high-scoring cases, where high score is indicative of suspected fraudulent activity on the payment card.During model training, we utilize the reason reporter algorithm to “bin” (sort) each input variable and then compute moments of the historical scores for each bin. In production, when a transaction is scored, the values of the variables are used to assign it to its bin. The smallest deviations of the score from the moments in each bin provide estimates of the input variables responsible for driving the score. Reason codes are then associated to a variable or group of variables. The schematic below of Falcon Case Manager, based on synthetic data used for demo purposes, shows a list of reasons (shown in the yellow box) generated by Reason Reporter for the highlighted transaction.
A Squeeze of LIME
As I wrote in my first XAI blog, perturbation-based approaches have recently gained attention due to a technique called Local Interpretable Model-agnostic Explanations (LIME). This technique involves manipulating data variables in infinitesimal ways to see which variables will change the score by the largest amount. LIME then fits a sparse model on the locally dispersed, noise-induced dataset. The indicator variables with the largest coefficients in the model are then reported as the drivers of the score.Working with LIME can be a challenge in that it explains local sensitivity vs. historical data support for reasons. At FICO, as we always do, we compared the newer algorithm to our existing methodologies.
For example, we studied a sample of high-scoring fraud cases, which we processed through both Falcon Reason Reporter and LIME. The Falcon neural network was used to generate the scores for the noise-induced perturbation data points required by the LIME methodology. The output of both algorithms was then qualitatively analyzed through expert evaluation. Among the high-scoring cases studied, 11% had exact same top three reason codes generated by Falcon Reason Reporter, and 53% had at two out of three of the same reasons codes.
In cases where the two systems differed, the story was interesting. For example, in a high-scoring case where both the algorithms agreed only on one of the three reasons, Falcon Reason Reporter pointed out that the high amount activity and authorization velocity were suspicious. On the other hand, LIME thought that the primary driver of the score was the rate of ATM withdrawal in the last two days, contrary to any evidence of the same.
Similarly in another case, with just a few transactions in the history, LIME thought that the reason for the score was PIN decline rate, though there were no PIN declines. We saw recurrence of such behavior again and again. These cases indicate that LIME over-emphasizes the local feature sensitivity induced through its local noise perturbation technique. On the other hand, by focusing on the global historical score patterns and global support, Falcon Reason Reporter is robust to the local noise.
Averaging Out Noise
The tried-and-true Reason Reporter’s behavior (looking at global score support in the historical data) is a superior method for averaging out noise caused in complex local variable phase spaces and gradients of the solution space. In contrast, the newer XAI technique (LIME) has a tendency to pick up the local noise, leading to contrived explanations made more complicated by increased non-linearity.Two decades after it was first introduced, Falcon Reason Reporter continues to provide robust and accurate explanations to set the investigation teams that utilize machine learning on the right track.
A Final Note
Most Explainable AI systems including Reason Reporter and LIME provide an assessment of which model input features are driving the scores. The ability to understand causality is the natural next step in the explanation systems. This requires the explanation system to understand and parse latent features that actually drive the score. As I wrote in my previous blog on this topic, my recent patent application work has explored an architecture called LENNS (Latent Explanations Neural Network Scoring) that exposes more of what’s driving the score. More on that later.If the robustness of the tech FICO originated in 1992 has got you dreaming of the '90s, check this out. And follow me on Twitter @ScottZoldi.
Popular Posts
Average U.S. FICO® Score at 716, Indicating Improvement in Consumer Credit Behaviors Despite Pandemic
The FICO Score is a broad-based, independent standard measure of credit risk
Read more
The GSE Data Is Public. Here's What It Shows About Credit Score Modernization.
Independent Milliman analysis of GSE historical data confirms what anyone can now verify directly: FICO® Score 10T is the most predictive score.
Read more
FICO® Score High Achievers: Is Age the Only Factor?
<p>FICO’s latest research on <a href="http://www.fico.com/en/blogs/risk-compliance/us-credit-quality-rising-the-beat-goes-on/" target="_blank">national FICO® Score distribution</a> shows that US credit quality continues to trend upwards, with an increasing number of consumers scoring in the highest scores ranges. Given that, we decided it was time to refresh our study on “<a href="http://www.fico.com/en/blogs/risk-compliance/infographic-what-does-a-high-credit-scorer-look-like/" target="_blank">FICO® Score high achievers</a>,” where we examine the credit behavioural profiles of consumers with higher credit scores.</p> <p>As expected, our latest study showed that older people generally have higher scores, as has been the case in our <a href="http://www.fico.com/en/blogs/risk-compliance/age-before-beauty-credit-worthiness-before-youth/" target="_blank">past research</a>. This is largely because they have trade lines that have been open longer, which leads to higher scores (assuming the trade lines are in good standing and all other things equal). The most credit savvy among you will remember that length of credit history accounts for roughly 15% of the overall <a href="http://www.myfico.com/CreditEducation/WhatsInYourScore.aspx" target="_blank">FICO® Score calculation</a>.</p> <p>But for younger consumers, it isn’t too helpful to say, “just wait until you are 50 years old, and that will fix things!” Thus, I decided to dive deeper with our analysis, including looking at the credit behavioural characteristics of younger consumers with high FICO® Scores.</p> <p>In this analysis, I looked at credit characteristics for the two highest score bands, 750-799 and 800-850, which have about 16% and 23% of the population, respectively. I further broke this down by looking at the characteristics of the youngest 15% in each group, which occurs at age 29 for the 750-799 group and 43 for the 800-850. We used FICO<strong>®</strong> Score 9 for our analysis, and examined typical variables of the categories that go into the make-up of the score.</p> <p><a href="http://www.fico.com/en/blogs/wp-content/uploads/2016/10/FICO-Score-High-Achiever-table.png"><img class="alignnone size-full wp-image-29367" src="http://www.fico.com/en/blogs/wp-content/uploads/2016/10/FICO-Score-High-Achiever-table.png" alt="fico-score-high-achiever-table" width="715" height="747" /></a></p> <p>Since there’s a lot to dissect in this chart, I’ll highlight a few findings that struck me: <ul> <li><strong>Credit history.</strong> Higher scores are undoubtedly associated with a longer time on file. What’s striking is the higher average time on file for the 800+ group, both overall and for revolving accounts (the most common trade line). Clearly, it helps to start credit use early in life – but that’s not the sole determining factor for high scoring. The younger cohorts in both the 800+ and 750-799 groups have time on file medians significantly less than their older counterparts.</li> <li><strong>New credit.</strong> High credit scorers do, in fact, pursue new credit. We see little variation in months since last trade line opening and last inquiry among all groups observed.</li> <li><strong>Pay on time.</strong> As you might expect, the further in the past the delinquency, the higher the score – and that’s <u>if</u> a delinquency ever happened. 95% of the 800+ people and 72% of the 750-799 group have never been delinquent, which is a striking contrast to the 57% for the total population. The percentages for the younger groups are even higher, especially so for the 750-799ers. Collection items are also scarce for these high scorers. While 23% of the national population have collection items greater than $99, the figures are only 0.5% for the 800+ group and 3% for the 750-799 group. It is clear that, in conjunction with starting a credit file early, paying on time is a key trait of the FICO® Score high achiever.</li> <li><strong>Balance and utilization.</strong> Finally, we look at how credit is being used, focusing on revolving product use. (Since most people only have one mortgage or one auto loan, variation in instalment product figures is minimal and driven by payment history, which is near perfect for high scorers, as we’ve seen.) FICO® Score high achievers tend to have more open revolving accounts. While the % active trades and utilization are generally lower, by no means do high scorers avoid the use of credit products. What drives the difference here is the amount of <u>available</u> revolving credit utilized. The 800+ group is roughly 1/4 of the national median, both in total and on their highest utilized card; they also have lower balances and more cards<em>. </em>The 750-799 group is more active and higher utilized, but still only at 2/3 of the national figures. The young 750-799 group compensate with lower revolving balances and, as a result, lower utilization, which is in line with what we see for the 800+ group.</li> <li><strong>Benchmarks.</strong> I’d like to point out that while this study provides good benchmarks of the typical credit traits of the FICO® Score high achiever, individual credit profiles will vary widely. And keep in mind that when we sort groups of consumers as a whole by different behaviours, some results may look superficially odd. For example, the 800+ band has a longer median average open account time (128 months) than average revolving months on file (140 months).</li> </ul> So, what do we conclude? A huge takeaway is that high scorers don’t have too much in the way of serious delinquencies and negative public record data.</p> <p>In most respects, the young 800+ people are like their peer group. They really only differ markedly in their shorter credit history and higher non-mortgage balance, both of which are a function of life stage. Our results show that you don’t have to be a credit product hermit; these high scorers have more credit cards. The takeaway is to start credit activity early and maintain well-managed and moderate usage.</p> <p>In many ways, the young 750-799 people are more interesting because their age profile is much younger. Their credit characteristics differ more from their peer group, particularly in terms of length of credit history. They compensate by having almost pristine credit records and lower balances.</p> <p>Ultimately, 39% of US consumers are FICO® Score high achievers, which they accomplish by following common-sense credit practices. It ain’t rocket science!</p>
Read moreTake the next step
Connect with FICO for answers to all your product and solution questions. Interested in becoming a business partner? Contact us to learn more. We look forward to hearing from you.