978-1-4799-7620-1/15/$31.00 ©2015 IEEE
Using Location Aware Business Rules for
Preventing Retail Banking Frauds
Ayhan Demiriz
Sakarya University
54187, Sakarya, Turkey
ademiriz@gmail.com
Betül Ekizoğlu
Sakarya University
54187, Sakarya, Turkey
ekizoglubetul@gmail.com
Abstract—Fraud detection procedures for national and
international economies have become quite an important task.
Ensuring the security of transactions carried out by banks and
other financial institutions is one of the major factors affecting
the reputation and profitability of such organizations. However,
since people who perform fraudulent transactions change their
methods constantly in order not to get caught up, it gets more
difficult to identify and detect this type of transactions. Detecting
this type of transactions makes the support of technology
compulsory, considering high volume and intensity of
transactions. In this paper, we explore practicality of using
location data to aid finding better business rules where they can
easily be deployed with a rule-based fraud detection and
prevention system for retail banking. In order to study the
importance of location data, we first compiled a set of
anonymized automated teller machine (ATM) usage data from a
mid-size bank in Turkey. Depending on how much mobile the
card owners are, we can easily devise business rules to detect the
anomalies. Such anomalies can be directed to appropriate
business units to be analyzed further or account owners may be
required additional authorizations for banking activities (such as
internet money transfers and payments). We have shown in this
paper that a significant bulk of ATM users does not leave the
vicinity of their living place. We also give some brief use cases
and hints regarding what types of business rules can be extracted
from location data.
Keywords—Location intelligence; fraud detection; sequence
mining; spatio-temporal outlier
I. INTRODUCTION
Financial fraud detection and prevention have been
receiving increasing attention in the past few years, due to the
dramatic increase of losses because of fraud transactions every
year. Fraud detection activities involves monitoring the
behavior of transactions and prevention means a proactive
approach that involves the analysis of transactions before they
completed and identifying if they are fraud or not [1]. In highly
connected societies that we now live in, financial fraud is very
common to the point that financial institutions form various
business units to guard their customers, capital and
infrastructure as well as their reputation. Fraud cases involve
criminal purposes and they are very hard to identify in most
cases. As new application channels increase in use (e.g.,
mobile), new fraud opportunities present themselves, and
anonymity becomes easier. The issue is that if financial
institutions’ fraud detection tools remain static, they can be
exploited by the fraudsters who quickly identify thresholds and
take advantage [3]. Financial fraud is composed of bank fraud,
securities and commodities fraud, insurance fraud and other
related financial frauds [4]. Detecting financial fraud is very
crucial for preventing often large scale and devastating
consequences [4].
Kou et al. present a survey of fraud detection techniques for
the types of credit card fraud, computer intrusion and
telecommunication fraud [5]. Phua et al. firstly defines the
fraudsters based on their motivation and identifies the affected
commercial industries from fraudulent activities in their survey
study [6]. They also formalize the main and sub types of fraud
and present the nature of data evidence collected within
affected industries. Based on the available data types they
examine supervised, unsupervised, semi-supervised and hybrid
techniques that are used for fraud detection.
Supervised and unsupervised methods can be used for
modeling fraudulent events [2]. Related predictive and
explanatory variables can be included in statistical, machine
learning and data mining models to understand, predict and
distinguish fraud from other events. Classification, outlier
detection, clustering, regression, and visualization are some of
the well-known methods used for detecting and preventing
fraud. Logistic regression, neural networks, Bayesian belief
networks, decision trees, and naive Bayes come in front of the
pack as the leading supervised methods used [4]. Clustering
and link analysis are major unsupervised methods utilized for
fraud detection and prevention.
In their survey studies about anomaly detection Chandola et
al. and Gupta et al. also reported that fraud detection problem
can be defined as detecting anomalies in discrete sequences
and the related data set can be classified as temporal or spatio-
temporal data sequences [7], [8]. Anderka et al. reported that
the total losses from ATM fraud during 2008 across Europe are
estimated to 485.15 million EUR. They propose a method for
ATM fraud detection in which they identify the ATM fraud as
a sequence–based anomaly detection problem [9].
On the other hand, behavioral profiles are very commonly
used in fraud detection systems [10], [11], [12], [13], [14]. In
their survey for financial fraud detection with signature based
methods Edge et al. provide a survey of existing research into
account signatures, fraud detection architectures, processing
models and applications. They summarized that the signature