Tutorial

Tutorial Topic: Using Data Mining to Detect Fraud in e-Banking Transactions

Dr. Tariq Mahmood
FAST-NU, Karachi

Targeted Participants: BS students, MS students, PhD students, Faculty, Representatives from Banking Sector

Details: The advent of e-banking has brought a corresponding risk for the financial sector. Many malicious users are intent on performing suspicious activities that can cause considerable losses to the bank itself, as well as to the other customers. Detecting such users, or malicious transactions, in advance is almost impossible with manual effort. Therefore, banks have recently started resorting to data mining techniques to indicate suspicious transactions well before they become officially authorized by the banks. In this work, we discuss the application of classification trees to detect different types of frauds from a dataset of Pakistani banks. We focus on account takeovers, card-not-present transactions, abnormal withdrawn amounts, abnormal amount of transfer, and transfer of funds to dummy accounts. We also employ time-series prediction techniques to predict  "failing" situations in Automated Teller Machines, e.g., wrong pin entry, machine-out-of-cash situation, a rare sequence of consecutive transactions etc. We are hopeful that our results are extremely valuable for the banking sector in Pakistan, and can assist them in saving large amount of money, along with enhancing customer relationship management.