Detection Method Of Mobile Electronic Payment Transactions In Abnormal Behavior

With the development of the Internet has also led to the development of e-commerce, including online payment technology has become a key technology to support e-commerce. But because the user base is growing, processing server for user data have become increasingly demanding, a lot of pressure to the server. Abnormal transaction detection model presented in this paper were from a time perspective of complexity and accuracy of analysis to compare the reduction of pressure on the server to verify the structural stability and sustain the conclusions presented in this paper a modified Bayesian network structure learning methods.This paper analyzes the shortcomings of basic Bayesian network structure learning algorithms scoring search method (K2algorithm and MCMC algorithms) exist, we propose a Bayesian network structure learning based on improved search algorithm score. The main research work and innovation of this paper is mainly reflected in the following areas:This paper analyzes the theory found under MCMC algorithm is a complete unknown, but the amount of data attributes in the topology of the network topology learning situation better solution, but found that the initial topology of the network topology learning methods for its network composed with great restraint, so training structure arising particularly vulnerable to produce local optima.K2algorithm to solve this algorithm concentrated in the topology of the network and data sets complete unknown topology premise training, which is a popular Bayesian network structure learning algorithms. However, after analyzing K2algorithm is the first prerequisite to get paternity order topological node, which meant that require expert intervention to proceed.According to the statements of the K2algorithm for MCMC algorithms and the pros and cons, I propose a MCMC algorithm based on K2and improved algorithms for learning Bayesian network structure, K2and MCMC methods are very dominant, I The average use of model theory with K2and MCMC theory study ways to improve the structure. The best part of the theory as a whole does not represent the best single theoretical calculations give the conclusion that there is an error in the trend, and this time the average theoretical models can spend, it can make the probability of error is small, so the fact that computing trends and more close, discover the best independent conclusions from theory into complete works best theory. Therefore, improved training methods Bayesian network topology first step to explore theories and models using MCMC methods to train the average order on the topology of the nodes, so that you can get the complete works best node order; then, has been put in nodes in order to use K2theoretical conclusions of the posterior probability of all nodes is calculated, and then compare the obtained optimal network topology.I deduced through improved algorithms and experimental results K2algorithm and MCMC algorithms contrast, verify my accuracy improved algorithm from the perspective of the analysis and time complexity of the structure of the quality and efficiency of learning algorithms and MCMC algorithms than K2higher (experience certificate, K2algorithm and MCMC algorithms compared with other traditional score search quality and efficiency of high learning whims).This method is simple to some extent solved by one process server stress factors as well as personalized intelligent identification data caused by various attributes, has practical significance.