Λογότυπο αποθετηρίου

Πλοήγηση ανά Συγγραφέα "Zafeiropoulos, Konstantinos-Efthymios"

A B C D E F G H I J K L M N O P Q R S T U V W X Y Z
Α Β Γ Δ Ε Ζ Η Θ Ι Κ Λ Μ Ν Ξ Ο Π Ρ Σ Τ Υ Φ Χ Ψ Ω
Φιλτράρετε τα αποτελέσματα πληκτρολογώντας τα πρώτα γράμματα
Τώρα δείχνει 1 - 1 από 1
  • Μικρογραφία εικόνας
    Τεκμήριο
    Anti money laundering mechanism for banking transactions
    (2022) Zafeiropoulos, Konstantinos-Efthymios; Ζαφειρόπουλος, Κωνσταντίνος-Ευθύμιος; Athens University of Economics and Business, Department of Informatics; Dellaportas, Petros; Demiris, Nikolaos; Vassalos, Vasilios
    The scope of thesis is to create a scalable anti-money laundering (AML) mechanism to detect transactions which can be considered as possible cases of money laundering and are carried out by National Bank of Greece (NBG) customers. Next, potential illegal transactions will be further evaluated by NBG authorized personnel. The main approach involves five main stages.First stage is data collection. All the transaction are not the same. There is plethora of different categories related to the nature of transaction. Hence, main emphasis on this stage was to find categories of transactions with a higher risk of money laundering appearance. The second stage involves feature extraction. Efficient money laundering mechanisms requires insightful features. Representative example is features based on monitoring different aspects of account behavior and its network. Abnormal behavior or network indicates higher probability of money laundering. Therefore, main emphasis on this stage was by utilizing the relevant theory to extract these types of features. The third stage is model selection. Given the fact that the specific task belongs to the sphere of unsupervised learning, the main idea is the creation of insightful features from the second stage to lead potential illegal transactions to be seen as outliers in vector space. Hence, main emphasis on this stage was by utilizing the relevant theory to choose unsupervised models which can detect outliers and simultaneously being computationally efficient. The fourth stage included a fine-tuning process upon different combinations of feature sets and models. Then, based on majority vote, scores were extracted. The final stage included the process of scores through visualization and statistical techniques which determined if the transaction was legal or illegal.