Anti-fraud system on the basis of data mining technologies

Abstract

The article is devoted to the issues of increasing the efficiency of transaction monitoring systems (TMS). The main types of TMS with their advantages and disadvantages are considered, and various algorithms of data mining used in this field were also considered. On the basis of the analysis of systems and algorithms, three classifiers were chosen: a multilayer perceptron (MLP), a classifier based on a random forest and a method of support vectors. The selected classifiers were tested on full-scale data. The best indicators were noted for the classifier on the basis of a random forest. The structure of the system for collecting and analyzing information about the user environment (UE) has been developed, which makes it possible to accumulate data about the UE, to mark use cases in manual and automatic modes and to build a database of images for training neural network classifiers. An algorithm for analyzing the information collected about the UE based on cluster analysis is developed.