FRAUD TRANSACTION DETECTION METHOD BASED ON SEQUENCE WIDTH DEPTH LEARNING

摘要

The present invention relates to a fraud transaction detection method based on sequence width depth learning, comprising: performing feature mapping processing on each of a plurality of transaction data, to generate corresponding feature vectors; converting, on the basis of a first self-learning model, the feature vectors of to-be-detected transactions into integrated feature vectors; converting, on the basis of a second self-learning model, the respective feature vectors of at least one timing sequence transaction into timing sequence feature vectors respectively; combining the integrated feature vectors and the respective timing sequence feature vectors corresponding to the respective timing sequence transactions, to form depth feature vectors; classifying the depth feature vectors on the basis of a third self-learning model, so as to determine that the to-be-detected transactions are normal transactions or fraud transactions. The present invention can efficiently improve the accuracy of a fraud transaction detection model detecting a fraud transaction.