Reported by Christina Emmanouilidou

One of the ways anti-fraud systems can utilise AI and machine learning is to support the detection of unusual patterns. AI and ML algorithms can sift through millions or even billions of transactions quickly to identify suspicious patterns or activities that might be signs of fraud. These algorithms learn to recognise new and changing threats, making them more effective at detecting new kinds of fraud.

How this works in practice:

Step 1 – Behavioural analysis: By using advanced analytics and machine learning algorithms, models are created that capture the typical behaviour patterns and relationships between different parties involved in transactions. This helps in understanding normal transactional behaviour and can be used as a baseline for comparison.

Step 2 – Identifying odd couples: Once the relationship patterns are established, it becomes possible to identify unusual connections or odd couples that deviate from the typical behaviour. These odd couples might include transactions between parties with no apparent connection, unusual transaction amounts or frequencies, or connections between high-risk individuals or businesses.

Step 3 – Investigating anomalies: When unusual connections or odd couples are detected, they are flagged for further investigation. This may result in transaction blocking, and in most cases will require review by compliance and risk management teams, supported by additional data enrichment performed from high-credibility sources about the parties involved in the transaction.

Step 4 – Refining models: As new odd couplings and unusual connections are discovered, the relationship patterns and models should be continuously refined and updated. This ensures that the detection systems adapt to evolving fraud tactics and maintain their effectiveness in identifying potential risks.

AI and ML algorithms are also utilised in the following anti-fraud capabilities:

• Risk Scoring: AI and ML models can give risk scores to transactions based on different factors, such as a user’s transaction history, account information, and behaviour patterns. Transactions with high-risk scores can be checked more closely or stopped, helping to prevent fraud before it happens.
• Understanding text: Natural language processing techniques should be used to analyse communication, like payment instructions and comments, to find possible anomalies but also to help establish the source of funds and clarify relationships and the purpose of the transactions.
• Login analysis: This detects suspicious logins by IP network providers outside the customer’s country by examining the transaction login history for unusual times and durations, such as after midnight and with a duration of under three minutes. This may include logins by several IP addresses and from different locations, as well as the common IP address logins of different customers.

Read full report: https://thefintechtimes.com/how-can-ai-be-used-to-thwart-financial-fraud-associated-with-real-time-payments-blackswan-technologies/