Eureka! Institutional Repository

This paper is about applying the most sophisticated data mining methods to a real dataset in order to detect possible fraud when opening a bank account. The contribution of this paper focuses on two areas. First, it attempts to highlight the strongest variables that are important indicators of possible fraudulent transactions. The identification, ranking by importance and evaluation of these variables is an important criterion for the acceptance or not of potential customers by a financial institution. Secondly, it suggests the creation of reliable models that can predict with high accuracy the majority of future cases. The paper is organized in three parts. The first part presents the role of financial institutions and in particular the function of the Compliance Unit in the fight against financial crime. It analyses the basic Know Your Customer principle and the institutional-legal framework currently in force in Greece. The definition of Money Laundering is given and the cycle of a money laundering scheme is described. Finally, the main data analysis technologies currently used to combat fraud are mentioned, as well as best practices that have been proposed for adoption in the future. The second part describes the stages of Knowledge Discovery, the four categories of Machine Learning and the most important supervised learning methods. Extensive reference is made to the specific topic of class imbalance, Cost Sensitive Learning and Performance Metrics of classifiers. The third part provides the methodology of this paper. To achieve the objectives, several methods are used. These include data pre-processing, feature selection and balancing the distribution of classes, developing models with machine learning algorithms, validating the models against unknown observations and finally evaluating and ranking in order of importance the variables leading to the generation of the models. Finally, the results of the analysis, conclusions, limitations and possible extensions for research are presented. At the end of each part, the relevant literature is listed.