Methods, systems, and computer program products for fraud prevention using deep learning and survival models

Abstract

A method of fraud prevention using deep learning and survival models is provided. The method may include receiving, with at least one processor, transaction data associated with a plurality of transactions of at least one payment account; at least one attempted attack may be detected based on the transaction data. A fraud risk score for each of a plurality of sub-periods in a time period following the at least one attempted attack may be generated based on the transaction data using a deep learning model and a survival model. The fraud risk score for each respective sub-period may be associated with a probability that the respective sub-period does not have a fraudulent transaction. A system and a computer program product are also disclosed.

Description

technical field [0001] The disclosed subject matter generally relates to methods, systems and products for fraud prevention, and in some specific embodiments or aspects, to methods, systems and computer program products for fraud prevention using deep learning and survival models. Background technique [0002] Service provider systems (eg, servers, etc.) in electronic networks can process a large number of events (eg, messages) on a daily basis. For example, a transaction service provider system in an electronic payment processing network may process thousands of transactions (eg, transaction messages, such as authorization requests and / or authorization responses) per second. Some of these transactions may be fraudulent, but determining which transactions may and / or are actually fraudulent may be difficult. For example, 100,000 to 300,000 account identifiers (eg, Primary Account Numbers (PAN), etc.) may be compromised per day. An attack may refer to an attempt to conduct a...

AI Technical Summary

Problems solved by technology

Additionally, such models cannot provide any indication of the likelihood of fraud at different times in the future, or account for changes in the likelihood of fraud at different times (e.g., sub-periods) within the time period

Furthermore, such models can only consider whether fraud occurred within a certain time period, not when such fraud occurred within said time period

In addition, it may be inefficient to run such models on a regular basis (e.g., daily, etc.), for example, due to the amount of time and / or resources it takes to run such models (e.g., to train and / or make predictions with such models)

Images