Research On Network Bank Abnormal Transaction Detection Technology And Application

Internet banking is an online electronic trading system built on Internet technology and information technology.How to use technical means to identify criminal acts against fund transactions such as transfer and electronic payment is a difficult problem for risk prevention and control management that banks and other financial management institutions are facing.At present,risk management models such as joint-stock commercial banks and other financial institutions have risks such as poor timeliness and high missed detection rate.This thesis focuses on the sub-module of bank transaction anti-fraud risk control based on deep learning,and applies and tuned deep learning in the abnormal transaction detection of bank transactions.The main contents of this thesis include the following aspects:1.A method of feature extraction of bank transaction data based on machine learning is proposed.Feature extraction is especially important for improving the fit of the model.The traditional feature extraction method uses Euclidean distance to measure the similarity of sample points,and can not measure the difference in the direction of space vector.This thesis proposes a feature extraction method that uses the similarity between sample space points of cosine metrics.2.An abnormality detection model for bank transaction data based on machine learning is proposed.The neural network of the machine learning model is solved by using the gradient descent algorithm to solve the local optimal solution.In this thesis,an improved learning algorithm is proposed,which uses dynamic learning rate update in the objective function solving process to solve the efficiency problem caused by the constant learning rate of the stochastic gradient descent algorithm.3.Develop online banking online real-time risk management system.Based on the machine learning model,the online banking abnormal transaction detection is realized,and the Spring Cloud micro-service framework is used to solve the horizontal expansion bottleneck of the single application.