Constructing a predictive model for Real-Time ATM CARD Fraud Detection
No Thumbnail Available
Date
2017-06
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
A.A.U
Abstract
Nowadays banks are getting closer to users through many channels. Due to the advancement of
technology ATM and POS channels are the most popular mode of payments; with the existence
of this technology the implication of financial crime against banks is accelerating rapidly. Fraud
prevention now represents one of the biggest areas of concern in financial institution. To detect
these types of attack in Ethiopia there is no precondition taken, though we are adopting many
Banking technologies every time.
Recognizing this problem, in this study an attempt has been made to create a predictive model
that helps to detect real-time ATM frauds using data mining techniques. Empirical research
methodology was used, and specifically the experiments were conducted following the
Knowledge Discovery in Database process model. The dataset used in this study has been
extracted from Bank ATM application server. A total of 51,500 records selected. The study was
conducted using WEKA software version 3.7.9 and four data mining algorithm namely, Simple
k-means to cluster the data set and J48, MLP, Naive Bayes algorithm used for classification.
Since the data set has poor data quality, data preprocessing was done on the original data set. The
major preprocessing activities include fill in missing values, remove outliers; resolve
inconsistencies, dimensionality reduction, size reduction and data transformation.
In this study the experimental result shows that ATM card fraud detection using K-means
clustering followed by J48 decision tree register the highest score of 90.8% of accuracy and user
acceptance testing shows that 91 % domain expert are satisfied with the developed model.
Finally the result of this study reveals that applying data mining for detecting novel type of
attack on ATM card related fraud generate interesting rules with the following attributes
(channel type, transaction type, transaction amount, location, Branch-location, fraud status,
terminal type). One major challenge was getting labeled data. Future research directions are
forwarded to come up with an applicable and reliable system in the area of this research by
having labeled training dataset and enable automatic synchronization from data mining software.
Description
A Thesis Submitted to the School of Graduate Studies of Addis Ababa University in Partial Fulfillment of
The Requirements for the Degree of Master
Of Science in Information Science
Keywords
ATM Card, Fraud Detection