Fraud detection often flags what looks unusual, not what is already known to be fraud
In short: Banks and insurers rarely wait for a fraud to be labelled before acting. Many systems flag observations that depart from normal behaviour, such as an order placed far from the cardholder's home. This approach can catch new kinds of fraud, though it also raises false alarms.
Anomaly detection means finding observations, events or patterns that differ markedly from expected behaviour. A flagged case is not always a data error. Depending on the setting, it might reflect a measurement fault, a shift in the process producing the data, a valid but unusual observation, or something of real interest such as fraud or a security intrusion. Sources describe three broad approaches. Supervised methods need data labelled as normal and abnormal, which is rarely available and is complicated by the imbalance between the two classes.
Semi-supervised methods build a model of normal behaviour and test how likely a new case is under it. Unsupervised methods work on unlabelled data and are by far the most common in practice. Many systems return a numerical score rather than a yes or no, and a chosen threshold turns that score into a decision. Whether something counts as anomalous therefore depends on the model of normality and where the threshold is set.
Credit card fraud shows these trade-offs. Online retailers and payment processors can compare a user's location with the billing or shipping address, and a mismatch is treated as a strong indicator of possible fraud. Card issuers use software that estimates the probability of fraud and may ask a merchant to verify or decline a transaction. One figure from the pack reports that in 2018 banks and card companies stopped 1.66 billion pounds of unauthorised fraud in the United Kingdom. Limitations remain. Link analysis, which relates known fraudsters to other people, mainly catches frauds similar to ones already classified, so new types may need unsupervised methods. Public datasets for testing fraud methods are scarce, which makes validation hard.
Opens the Intellect+ feed with this card first. Swipe up for the next one.
Sources
- Summary by Intellect · source: Anomaly detection
- Also: Data analysis for fraud detection - Wikipedia
- Also: Credit card fraud
To cite this card, link to https://intellectplus.co.uk/app/wealth-fraud-detection.
How we source, check and correct cards: Editorial standards.
General information, not professional advice. Spotted an error? Flag the card or see corrections and complaints.
Intellect Plus is a visual educational discovery platform for exploring fascinating facts and ideas. Brain Food, Not Brain Rot. What is Intellect Plus?
Related
More in Researched · More in Wealth & Business · Latest · A–Z · All topics