How to use Artificial Intelligence for fraud detection
2 min read
The possibilities for using Artificial Intelligence in daily life are vast, ranging from meeting analysis and accelerating software development to fraud detection. This latter application was implemented by our team a few years ago, benefiting one of our clients.
The service provider’s challenge was to reduce the acts of bad faith it had been experiencing. The CWI team then developed a solution that uses AI to evaluate anomalies in the services provided and identify suspected fraud.
Continue reading to learn more about the tools we used – they might be useful for your situation as well!
Machine learning for an ever-improving fraud detection system
The starting point of the project was identifying which data could form the solution. Information about recommended products, for example, could indicate some kind of exaggeration. Similarly, examining the times of service records could reveal some inconsistency.
The Strategic Software Engineering team began the work with the goal of identifying potential fraud scenarios so that our client could intervene before they materialized. Machine learning, a subset of Artificial Intelligence, was the chosen approach for the project. This way, the system could improve fraud detection over time based on experience.
Previously, the data was stored in a relational format. However, this configuration does not favor use in machine learning models. Therefore, the data was processed to become useful for machine learning techniques.
Statistics in a service of Artificial Intelligence
In the solution developed by CWI, data analysis was conducted in various ways. One of them utilized the concept of z-score (standard score), a statistical technique that relates a value to the mean of a group of values. In other words, it quantifies how much a value deviates from the mean in terms of standard deviation.
This approach applies when standardizing and comparing data within a distribution, such as in the case of medical appointment scheduling. In cases where there is a sequence, our time has observed that the schedules are interconnected and have patterns in the period between schedules and the time of day in which they occur.
As it makes sense to analyze them together, the team then identified the z-score of the appointment times and the z-score of the time intervals between appointments. This approach allowed for making the variables comparable to each other and identifying anomalies, thus enabling fraud detection.
The solution contributed to saving resources for our client. The theory behind it can be applied in many other contexts, get in touch with us to explore ways to use Artificial Intelligence to optimize results!
