Before embarking on an AI initiative, companies must understand which technologies perform what types of tasks, and the strengths and limitations of each. Rules-based expert systems and robotics process automation, for example, are transparent in how they do their work, but neither are capable of learning nor improving on fraudulent behaviour strategies.
Deep learning, on the other hand, is excellent at learning from large volumes of data, but it’s almost impossible to understand how it creates the models it does. This issue can be problematic in highly regulated industries such as financial services, in which regulators insist on knowing why decisions are made in a certain way.
In the next few Blogs, the 5 keys to using AI and machine learning in detecting fraud in the South African Micro-Insurance sphere will be discussed. The insights here are based on XTND’s experience in fraud risk mitigation solutions, with specific focus on the management and support of high volume, quick-turn-around-time claims processes.
Integrating Supervised and Unsupervised AI Models in a Cohesive Strategy
Because organized crime schemes such as funeral fraud schemes or hospital cash-back fraud schemes are so sophisticated and quick to adapt, defense strategies based on any single, one-size-fits-all analytic technique will produce sub-par results. Each use case should be supported by expertly crafted anomaly detection techniques, that are optimal for the problem at hand. As a result, both supervised and unsupervised models play important roles in fraud detection and must be woven into comprehensive, next- generation fraud strategies.
A supervised model, the most common form of machine learning across all disciplines, is a model that is trained on a rich set of properly “tagged” transactions, also referred to as a claims data set. Each transaction is tagged as either fraud or non-fraud. The models are trained by ingesting massive amounts of tagged transaction details to learn patterns that best reflect legitimate behaviours. When developing a supervised model, the amount of clean, relevant training data is directly correlated with model accuracy.
Unsupervised models are designed to spot anomalous behavior in cases where tagged transaction data is relatively thin or non-existent. In these cases, a form of self-learning must be employed to surface patterns in the data that are invisible to other forms of analytics. Unsupervised models are designed to discover outliers that represent previously unseen forms of fraud. These AI-based techniques detect behavior anomalies by identifying transactions that do not conform to the majority. For accuracy, these discrepancies are evaluated at the individual level as well as through sophisticated peer group comparison.
By choosing an optimal blend of supervised and unsupervised AI techniques, you can detect previously unseen forms of suspicious behavior, while quickly recognizing the more subtle patterns of fraud that have been previously observed across thousands of claims data.