|
...billions dollars worldwide (Ghosh and Reilly, 1994; Aleskerov et al., 1997). The quick detection of fraudulent behavior to avoid such losses has become critical in a variety of industries. In particular, many service industries are using versatile approaches to detect fraud in various kinds of customer activities (Chan and Stolfo, 1998; Emran and Ye, 2002; Ye, Ehiabor and Zhang, 2002). A comprehensive review of fraud detection via statistical techniques has been published by Bolton and Hand (2002).
In the telecommunications industry, for example, it is estimated that more than 200 variants of telecom fraud exist and that this number is growing with the advent of new services (Jacobs and Booth, 2000). Detection of telecom fraud can be challenging for the following reasons among many others. One is the existence of complex patterns in the customer records due to the business dynamics of each individual customer and variations among customers. Telecommunications customer records usually include multivariate attributes, such as a customer's usage, originating and destinating phone numbers, network block status, etc. These records are often contaminated by seasonality and/or holiday effects and the impact of marketing campaigns, which increase the difficulty of analysis and detection. Moreover, although profiling methods have been developed to track customers' dynamic behavior, it is not easy to detect unusual patterns at an early stage before the data set is complete. For example, customer usage patterns may be analyzed routinely based on weekly customer records to detect suspected fraudulent behavior. However, to reduce fraud losses, it is much more desirable to perform such an analysis earlier in a week, say on Wednesday.
Coincidentally, in manufacturing statistical analysis, fault detection in batch processes faces similar problems. Manufacturing batches usually generate complex data patterns and correlations due to a combination of within-batch variation and batch-to-batch variation. The multivariate nature of the data also increases the challenge in analysis. Moreover, it is important to detect an abnormal batch as quickly as possible before production is complete. Fortunately, manufacturing batch analysis and fault detection is a relatively mature area. Therefore, based on the similarity of these areas, we suggest that some well-developed batch process control techniques in manufacturing industry may be able to be applied to fraud detection problems in service industries. This paper will propose modified manufacturing batch techniques which can help to detect frauds when customer information is incomplete.
The remainder of this paper is organized as follows. Some existing methodologies to handle fraud detection and batch processes will be reviewed. A method modified from batch manufacturing will then be proposed to deal with the fraud detection problem in the telecommunications industry. After that, a real case from a telecommunications company will be used to demonstrate the efficiency and economic performance of the proposed method. The last section will conclude this paper by setting out some issues for further research.
2. A review of related work
Fraud detection is important in various businesses, such as lending, telecommunications and financial services. As a result, effective methods have been developed to address the problem (Bolton and Hand, 2002). Statistical fraud detection methods can be divided into supervised and unsupervised methods. In supervised methods, models are constructed based on a historical data set with known fraudulent and non-fraudulent cases. Using defined criteria and threshold values, a new observation can be assigned as being either fraudulent or non-fraudulent. Widely used statistical classification methods such as linear discriminant analysis and neural network methods have been shown to be effective tools in such supervised situations (Hand, 1981, 1997; McLachlan, 1992; Ripley, 1996; Webb, 1999). On the other hand, unsupervised methods try to find those observations that are most different from a norm. A unified profiling and outliers detection approach is often used in fraud detection methods based on checking for suspicious changes in user behavior (Hill, 1996; Nigrini and Mittermaier, 1997; Murad and Dotas, 1999). However, the detection of individual frauds may require methodologies tailored to the individual cases. Ye, Borror and Zhang (2002) and Ye and Chen (2003) used exponentially weighted moving average control chart techniques to highlight different types of data and event changes in computer intrusion detection. Also, statistical data mining techniques are used in credit card fraud detection (Brause et al., 1999) and telephone calling fraud detection (Cox et al., 1997).
[FIGURE 1 OMITTED]
In manufacturing, a batch process is one in which raw materials go through a process with predetermined start and end points (Morris and Watson, 1997). Mason et al. (2001) proposed the use of a Statistical Process Control (SPC) method, the Hotelling's [T.sup.2] chart, to batch processes with phase I and phase II operations, so that outliers are detected and removed in the phase I operation, and future observations are collected and monitored in the phase II operation (see Montgomery (2001) and references therein for a general introduction to SPC).
Figure 1 depicts a typical approach to handling a batch process (MacGregor et al., 1994; Nomikos and MacGregor, 1995; Cho and Kim, 2003). The three axes of the top left shape in the figure represent the variables describing each batch, different batches and different times. By unfolding the original data, this three-dimensional data can be transformed into two dimensions, as in...
NOTE: All illustrations and photos
have been removed from this article.
Looking for additional articles?
Search our database of over 3 million articles.
Looking for more in-depth information on this industry?
Search our complete database of Industry & Market reports by text, subject, publication
name or publication date.
About Goliath
Whether you're looking for sales prospects, competitive information, company
analysis or best practices in managing your organization,
Goliath can help you meet your business needs.
Our extensive business information databases empower business
professionals with both the breadth and depth of credible,
authoritative information they need to support their business
goals. Whether it be strategic planning, sales prospecting,
company research or defining management best practices -
Goliath is your leading source for accurate information.
|
