Scientists from Perm have introduced a new method to combat online fraud.

E-commerce refers to the realm of electronic commerce, encompassing online trading and financial processes. These transactions are often targeted by fraudsters due to the convenience they offer for stealing credit card information and making purchases without direct interaction with the seller. Banks employ specialized software known as anti-fraud systems, which analyze transactions to detect unusual purchasing behavior and take appropriate action. However, cybercriminals continuously exploit vulnerabilities, and updating systems to block new methods of digital theft requires significant time, manual effort, and extensive data analysis. Researchers at Perm Polytechnic University have developed self-learning software that reduces the risk of fraudsters circumventing control systems.

2025-02-21 10:00:11https://naked-science.ru/article/column/moshennichestvom-v-intern

The article has been published in the collection of student conferences "ASUIT." The research was conducted as part of the strategic academic leadership program "Priority-2030."

ECOM transactions are the most common type of operation for fraud. The fraud rate (obtaining financial benefits through deception) in such transactions reaches three percent of the total volume. This surpasses the level and frequency of fraud in other types of transactions: intra-bank (0.1 percent), interbank transfers (0.5 percent), and card payments (0.1 percent).

To address this issue in the domestic market, fraud monitoring systems have been implemented. However, existing programs lack multi-stage checks, causing the detection rate of fraudulent activities to decline each year.

Scientists at Perm Polytechnic University are developing an anti-fraud program with three different and independent modules. These systems will internally build models and learn. Each will include three main methods: clustering, classification, and decision tree algorithms.

“User behavior is monitored by systems on servers: the days and times of online transactions, amounts, and so on. This creates a so-called 'acceptable area' into which each operation should fit. Clustering and classification are the initial stages of detecting fraudulent activities, which, aided by algorithms, analyze the transaction and determine whether it deviates from the user's usual behavior,” comments Alexander Subbotin, a graduate student at the Department of "Information Technologies and Automated Systems" at PNIPU.

“Next, after these two stages, if the system cannot determine whether the user initiated the payment independently, the data is analyzed using the decision tree method. The system module evaluates the parameters of the operations and forms an appropriate verdict based on them. The algorithm of the proposed anti-fraud system includes the parallel operation of three independent modules, each containing three stages of verification.

This multi-stage process increases the likelihood of detection compared to systems with a single module. Currently, the software is in the process of gathering statistics and training, and in the future, it could reduce the risk of fraudulent transactions, raising the detection rate to 95 percent and above,” explains Rustam Faizrakhmanov, a professor and head of the "Information Technologies and Automated Systems" department at PNIPU, Doctor of Economic Sciences.

The development by PNIPU scientists could be used to detect and prevent digital fraud during online purchases. The proposed algorithm and the use of artificial intelligence will eliminate the possibility for offenders to bypass such a system, as it will self-learn and adapt to changes and the recognition of new illicit activities.