Supervised and unsupervised learning techniques for fraud detection.

Artificial intelligence is redefining fraud prevention techniques as it allows to obtain information based on experience. This information consists of transaction activities, behavior and trends. Before the use of artificial intelligence, the applied methods were based on rules that helped to analyze historical fraud patterns but could not prevent them. Although these models could identify fraud attempts, they did not provide information of the future.

Nevertheless, the technological sophistication of fraud crimes is higher and more precise and efficient attacks have grown during the last few years

As a result, leading companies dealing with potential fraud crimes, mainly in banking and insurance, must increase monitoring accuracy and acuity of customers potential risk for the institution. Decision-making on accepting or rejecting payments, limiting charges refund and reducing operational and reputational risks is much easier now.

Fraud prevention in the future will depend on a combination of supervised and unsupervised automatic learning techniques. Supervised automatic learning finds patters based on historical events, factors, trends, etc., and unsupervised automatic learning looks for relationships and variable links, a combination of both methodologies would help to prevent fraud in the following:

Detection in real time. The use of artificial intelligence enables the detection of attacks in real time, instead of weeks that usually takes to start receiving reverse requests of charges.

Thwarting the most sophisticated attempts of fraud. Fraud techniques get more and more sophisticated. Artificial intelligence would help to prevent and reduce these attacks.

Scoring in real time. Provide analysts a scoring for a better perspective to set the limits to maximize sales and minimize losses in real time.

Immediate transactions. Fraud prevention systems based on AI enable immediate transaction´s approval provided it is within the reverse charges threshold of the main debit and credit cards.

Reduction of false positives. False positives are reduced thanks to supervised and unsupervised automatic learning, whereas current techniques cannot efficiently detect them. Frequently, when a customer pays an unusual amount of money or from a new location, the card is blocked by the system as it interprets it wrongly as a suspicious activity. With artificial intelligence, it is possible to identify more precisely any change in customers expenditure habits.

Profitability in low margin products. AI has allowed insurance companies to continue their profitable business and attract new customers whose historical purchases are not part of the historical supervised learning of fraud systems

Supervised and unsupervised learning should be complemented with experts´ knowledge aiming at a mixed approach to focus their attention in more suspicious cases detected by AI.