Λογότυπο αποθετηρίου

Πλοήγηση ανά Συγγραφέα "Zeakis, Alexandros F."

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
    Anomaly detection on system logs with graph analytics
    (2018-12-21) Zeakis, Alexandros F.; Athens University of Economics and Business, Department of Informatics; Andritsaki, Konstantina; Kotidis, Υannis
    The goal of this thesis is to study data provided by the National Bank of Greece andcreate a graph representation, in an effort to detect anomalies. In doing so, we hadto think of an efficient way of representing all the entities and relationships from anRDBMS into nodes and edges. Then, we used the Neo4j graph database toaccommodate the graph and Cypher to execute the corresponding queries thatdetected the anomalies. As anomalies we indicated transactions and accounts,which were merely suggestions that the bank would confirm or deny after the end ofthis thesis.The innovation that we offered NBG through this thesis was that we utilized data,that were never used in the past, due to the fact that they were semi-structured in anSQL environment and that we tried to analyze and exploit anomalies throughcommon graph analytic algorithms, while also offering some insights about theirclients by ranking them with graph metrics such as PageRank and ClosenessCentrality.The analysis consisted of two parts: in the first part we defined and calculated graph& monetary metrics, that created a description for each account and we suggestedan analysis workflow with statistical outliers, where we used these metrics to findtransactions that seemed as anomalies. In the second part, we studied specificuse-cases, such as Money Laundering or Fictitious Accounts, trying to find chains oftransactions or suspicious accounts, accordingly. In the first part we examined eachaccount separately and found individual anomalies, while in the second part westudied subsets of the graph, resulting in suspicious behaviours due to groupedinteractions.As a result of this process, we offered the bank specific accounts that should bemonitored and transactions that were flagged as possible anomalies.