Fraudulent transaction identification method, system and storage medium based on deep learning

Abstract

The invention discloses a method, system and storage medium for identifying fraudulent transactions based on deep learning. The method includes: acquiring training samples, which are transaction data for establishing a fraudulent transaction detection model; constructing a stacked restricted glass Erzmann machine RBM neural network structure and training, and carry out dimensionality reduction and clustering to described training sample by the RBM neural network structure of training, to divide training sample into several groups; Calculate the centroid of all groups, and calculate the Hamming distance between each group and the centroid respectively; determine the fraud probability of each group according to the calculated Hamming distance to establish a fraudulent transaction detection model; obtain transaction data to be detected, and according to the fraudulent The transaction detection model analyzes the transaction data to be detected to obtain the fraud probability of the transaction data to be detected, thereby identifying fraudulent transactions. The invention can improve the accuracy and rationality of identifying fraudulent transactions.

Description

technical field

[0001] The present invention relates to the field of financial risk control, in particular to a method, system and storage medium for identifying fraudulent transactions based on deep learning. Background technique

[0002] The financial sector has higher requirements for transaction risk control. In the identification of fraudulent transactions using deep learning, currently supervised learning algorithms are generally used to train detection models, and the features used to train detection models are constructed based on labeled historical transaction data, so the detection models trained by supervised learning algorithms , can effectively identify historical fraud types, but generally can’t do anything about unknown fraud types that lack fraud samples (such as fraudulent transactions that have never appeared or have variants). This posteriority leads to lag in transaction risk identification and low accuracy.

[0003] On the other hand, the existing metho...

Examples

Embodiment Construction

[0054] It should be understood that the specific embodiments described here are only used to explain the present invention, not to limit the present invention.

[0055] The main solution of the embodiment of the present invention is to: obtain training samples, which are transaction data used to establish a fraudulent transaction detection model; construct a stacked restricted Boltzmann machine RBM neural network structure and perform training, and pass The trained RBM neural network structure performs dimensionality reduction and clustering on the training samples to divide the training samples into several groups; calculate the centroids of all groups, and calculate the Hamming distances between each group and the centroids respectively ; Determine the fraud probability of each group according to the calculated Hamming distances to establish a fraudulent transaction detection model; obtain the transaction data to be detected, and analyze the transaction data to be detected ac...

technical field

[0001] The present invention relates to the field of financial risk control, in particular to a method, system and storage medium for identifying fraudulent transactions based on deep learning. Background technique

[0002] The financial sector has higher requirements for transaction risk control. In the identification of fraudulent transactions using deep learning, currently supervised learning algorithms are generally used to train detection models, and the features used to train detection models are constructed based on labeled historical transaction data, so the detection models trained by supervised learning algorithms , can effectively identify historical fraud types, but generally can’t do anything about unknown fraud types that lack fraud samples (such as fraudulent transactions that have never appeared or have variants). This posteriority leads to lag in transaction risk identification and low accuracy.

[0003] On the other hand, the existing metho...

Examples

Embodiment Construction

[0054] It should be understood that the specific embodiments described here are only used to explain the present invention, not to limit the present invention.

[0055] The main solution of the embodiment of the present invention is to: obtain training samples, which are transaction data used to establish a fraudulent transaction detection model; construct a stacked restricted Boltzmann machine RBM neural network structure and perform training, and pass The trained RBM neural network structure performs dimensionality reduction and clustering on the training samples to divide the training samples into several groups; calculate the centroids of all groups, and calculate the Hamming distances between each group and the centroids respectively ; Determine the fraud probability of each group according to the calculated Hamming distances to establish a fraudulent transaction detection model; obtain the transaction data to be detected, and analyze the transaction data to be detected ac...

Images