Mathematical methods of financial transaction evaluation for fraud

A n increase in the number of the financial transaction is currently observed, which triggers more financial frauds and more losses from the cyber attacks in the global economy. Detection of the deviant transactions is a burning issue for modern studies because all bank system participants are looking for minimizing the risks which could arise from the vulnerabilities in online transaction. An increase in the financial losses caused by the financial fraud updates the importance of the mathematical methods to analyze the real data. The purpose of the present study is to develop and to define the best mathematical model to predict fraudulent transactions. The novelty of the study lies in designing different binary choice models based on the panel data to predict the deviant transactions, as well as to compare the econometric models with the models based on the neural networks and tree ensembles and in justifying the choice of the best model. Methodologically, the study applies correlational analysis methods, econometric and neural network methods, decision tree ensembles. The most significant results referred to the scientific novelty of the research are as follows: 1) panel data-based financial transactions have been econometrically analyzed within probit- and logit-models with fixed or random effects; 2) neural network methods and tree ensemble-based method have been applied to predict fraudulent transactions; 3) designed mathematical models have been comparatively analyzed, and the model giving the best result in detecting the fraudulent transaction has been defined. Further research is connected with more profound study of the impact of different factors to check the financial transactions for their fraud nature.