Classification adversarial fraud detection method based on class clear representation

Abstract

The invention discloses a classification adversarial fraud detection method based on class clear representation, which adopts an improved auto-encoder and an improved generative adversarial network, and comprises four stages: a first stage: extracting preliminary features based on normal behavior data, and extracting data features thereof by using the improved auto-encoder; a second stage: modifying a target function of the original generative adversarial network to obtain an improved generative adversarial network, and generating pseudo-abnormal behavior data by using the improved generative adversarial network; a third stage: inputting the normal user behavior data and the pseudo-abnormal behavior data into an improved auto-encoder for training together, and extracting final features from the normal behavior data by using the trained encoder; and a fourth stage: training the improved generative adversarial network by using the finally extracted normal behavior data features, and detecting fraud by using a discriminator obtained after training as a fraud detector. According to the method, the accuracy and the stability of fraud detection are remarkably improved.

Description

technical field [0001] The invention relates to the technical field of data processing, in particular to a classification-based adversarial fraud detection method based on explicit representation of classes. Background technique [0002] Fraud exists widely in our lives, in network, telecommunications, insurance (health, car, etc.) claims, banking (tax declaration claims, credit card transactions, etc.) major problem. [0003] Technically, fraud detection mainly has the following technical problems: [0004] (1) Frauds are generally adaptable and changeable. Some traditional methods are based on robust features manually extracted by experts for detection. Once the type of fraud changes, the previous methods are no longer applicable. [0005] (2) The data volume and data dimension of fraudulent behavior are large. With the increase of data volume and data dimension, the calculation amount of traditional methods increases exponentially, and the calculation cost is too high...

AI Technical Summary

Problems solved by technology

Some traditional methods are based on robust features manually extracted by experts for detection. Once the type of fraud changes, the previous methods are no longer applicable

[0005] (2) Fraudulent behavior data volume and data dimension are large

With the increase of data volume and data dimension, the calculation volume of traditional methods increases exponentially, and the calculation cost is too high

[0006] (3) Data on fraudulent behavior is not easy to obtain

We have seen that although OCAN solves the problem of lack of negative samples with a generative network, and extracts features from samples to improve the ability to process data, a good feature extractor should maximize the inter-class distance and minimize the intra-class distance as target, so the feature extractor trained only by normal behavior data will ignore the target of maximizing the distance between classes, resulting in deviation

Images