Аналіз різнорідних даних в інтелектуальних системах виявлення шахрайства

Abstract

Fraud is being considered as an anomaly in the data in the work. The work is devoted to the development of a method of heterogeneous data analysis in intelligent fraud detection systems. The detection of fraud as an anomaly in heterogeneous data during mobile applications installation using set theory, which allowed further data analysis in such systems, is formalized. The mathematical model of the process of heterogeneous data analysis, the algorithm of heterogeneous data analysis, the method of heterogeneous data analysis on the basis of the proposed scales and coefficients that allowed processing of various input data — data of various metrics, templates, dimensions, which in the analysis process makes it possible to form a generalized fingerprint of fraudster, is proposed. The developed method uses the databases and knowledge bases, through which a generalized fingerprint of the fraudster is formed, the presence of which allows accelerating the detection of fraudsters in new data sets and detecting even implicit fraudsters. The proposed method is designed to use it in intelligent systems for fraud detection based on anomalies in data that, unlike existing ones, will allow analyzing heterogeneous data on the basis of which fraudulent decisions are made, to reduce the dimensionality of data and to classify users. Experi-mental researches of the proposed method of heterogeneous data analysis as a part of detection of fraud as anomalies in heterogeneous data and a classification model developed using fully connected deep neural networks with three hidden layers using the developed software and using a representative sample have been carried out. The scheme of experimental research of detection of anomalies in heterogeneous data during the mobile applications installation, based on which meth-od of heterogeneous data analysis was presented has been proposed. The efficiency of using the proposed method in the fraud detection system is shown, the classification accuracy of which was 99,14 %, the accuracy of the fraud detection is 82,76 %. However, with the increase of rules in the developed knowledge base, which will grow with each launch on the new data, the accuracy of the system will increase.