Artificial Intelligence (AI) delivers improved telecom fraud detection capabilities to identify fraud, and at the same time, empowers telcos to automatically adapt to new fraud scenarios in wholesale telecommunications.

Typical telecom fraud platforms are rules-based systems. These platforms can capture many fraudulent instances but have limitations due to the effort required to manage rules that achieve a sophisticated level of complexity. Conversely, AI based anomaly detection, learns the complexity inherent in the data. We discuss in this article how AI can deliver superior capability to identify fraud, as well as the ability to continue to adapt as the fraudsters adapt.

Why Telcos Must Prioritise Fraud Prevention

Telecoms fraud is on the raise. All telco operators are open to potential fraud attacks 24/7. With cheaply available telecoms equipment and information and support available from communities on the dark web, fraudsters can quickly generate revenue streams at the expense of telcos and their customers. The trend in recent years is an increase in the frequency and sophistication of these attacks, and consequently, an increase in the losses incurred by carriers worldwide.

The Role of Wholesale Fraud Management Systems

The typical response to these attacks is the deployment of a fraud detection system, usually based on analysis of data from signal record monitoring or call detail records (CDR). Traditionally these systems rely on a rules-based method for the identification of telecom fraud. The data is analysed or aggregated, and validated against a set of rules to find potential fraud cases. These cases are then either passed to an analyst to assess, or can be employed as a trigger for automatic blocking to prevent the fraud continuing.

There is a shortcoming inherent in this rules-based approach. The more sophisticated its capabilities become, either by increasing the number of rules used or increasing a rules complexity, the greater the difficulty in the management of rules. Resources and tools are required to analyse and amend the set of rules to successfully identify telecoms fraud, while keeping false positive cases to a minimum. In addition to this, the identification of fraud is a moving target, the current normal patterns will not persist, where customer behaviours, call volumes and attack patterns change over time.

Leveraging Artificial Intelligence for Fraud Detection

An approach that avoids the need for these rules, is to use a fraud management system based on machine learning (ML) and artificial intelligence (AI). This is a modern approach for fraud detection, leveraging the technology’s ability to learn the complex patterns inherent in telecoms data and to automatically create system generated rules. However, in practice, there are some difficulties in the application of ML that need to be considered.

The typical ML approach, known as supervised learning, relies on what is referred to as labelled data. Labelled data in this case, means call information that is labelled as either fraud or non-fraud. Supervised learning can then learn the patterns in the data that result in instances of fraud. It may be possible to acquire some labelled data with the help of several analysts, but this will prove to be an expensive and laborious process, given that ML systems are hungry for data. Additionally, because the quantity of telecom fraud cases will be such a tiny proportion of the overall data, known as a highly imbalanced dataset, the learning process will be very challenging. Therefore, any supervised learning techniques will very likely perform poorly, if at all.

Anomaly Detection Systems in Action

AI based anomaly detection systems can capture the highly complex informational structure within the data, to be able to learn the numerous call patterns that typically occur. Any calls that appear to deviate from these patterns are flagged as atypical, and can be assessed by an analyst. Once enough anomalies have been analysed and classed as fraud or non-fraud, we will have produced labelled data that can then be used to extend the anomaly detection system using supervised learning. This can be expanded to identify the particular telecom fraud type that is occurring, e.g., PBX hack. We then have a system that can identify and classify previously seen fraud types, and will flag as anomalous any new unseen occurrences which can be analysed and used to train the system further.

Wholesale Telecom Fraud Detection Solutions with iCONX

iCONX Fraud management system is driven by artificial intelligence to combine both inbound and outbound voice traffic analysis. Using the latest technology and industry insight, we are expertly placed to help you combat wholesale telecoms fraud for international voice traffic on your network. Contact iCONX today to discuss how we can help you with your wholesale fraud management needs.

About the Author

Jonathan Keaveney is head of software development in iCONX, and has been part of our team since 2010. He has been working in the technology sector for over 20 years, and has significant expertise in building enterprise applications for the telecoms sector. As head of software development, Jonathan fosters best practices within our software development team, and continuously adopts modern tools and platforms to assure quality solutions are delivered to our customers.

Jonathan has recently graduated with a Master of Science in Artificial Intelligence from the University of Limerick. Jonathan and his teams are looking at even further ways to advance the iCONX wholesale product set using AI technology, and to support next generation services for our telco customers.