Infosys Card Fraud Detection

Overview

Credit card fraud remains a persistent challenge for the industry, despite security advancements like multifactor authentication, tokenization, and biometrics. Globally, credit card fraud has seen a significant 46% year-on-year increase, with losses projected to reach $43 billion by 2026, of which the US market alone will account for $12.5 billion. Fraud rates for in-person transactions (CP) stand at 0.06% of the transaction value, while card-not-present (CNP) transactions face a higher fraud rate of 0.93%, which is 90% higher.

Banks and card issuers employ various internal or third-party commercial off-the-shelf (COTS) products to combat fraud. Most of these fraud protection solutions rely on rule-based engines to predict potential fraudulent transactions based on past customer behavior and transaction parameters. Recently, there has been a shift towards AI-based solutions leveraging advanced algorithms and increased computing power to enhance fraud prevention measures.

Utilize AI algorithms not only to detect fraudulent credit card transactions but also extract actionable insights through analytics, thereby strengthening fraud prevention mechanisms

Talk to our experts

Infosys Card Fraud Detection employs AI algorithms for swift detection of fraudulent credit card transactions. Beyond detection, it harnesses advanced analytics to derive actionable insights, bolstering fraud prevention efforts.

Features of Confusion Matrix

The confusion matrix summarizes the test results of any fraud management system:

True positive

Both the expected and predicted values are positive (1-Yes), indicating that the model accurately identified a fraudulent transaction.

False positive

The model incorrectly predicted a positive outcome (1-Yes) or a fraud, when the actual value was negative (0-No), i.e., it was a genuine transaction which often leads to customer dissatisfaction.

True nagative

Both the expected and predicted values are negative (0-No), indicating that the model correctly identified a non-fraudulent transaction.

False negative

The model incorrectly predicted a negative outcome (0-No) when the actual value was positive (1-Yes). In other words, it was actually a fraudulent transaction which can potentially result in financial losses for the bank.

Desired Outcomes and Minimization of Errors

True positive and true negative are desirable as they reflect accurate predictions. Conversely, banks aim to minimize false positives and false negatives to prevent customer dissatisfaction and financial losses, respectively.

This solution helps banks analyze test results across multiple models and approaches. Banks can determine their line of defense based on the following parameters:

Improved recall: Fewer false negatives
Balance with precision: Minimize customer dissatisfaction and the cost of manual interventions
Evaluate cost/loss: Assess the financial impact of missed fraud alerts

When evaluating models, banks must decide on the tradeoff between prioritizing precision or recall.