Investigative Data Mining Articles

Data mining provides an incredibly powerful tool
for all those interested in the identification of
malpractice and, as a result, the exacting
needs of detecting the symptoms of fraud
have continued to push data mining techniques to their limit. There are also, however, many commercial applications, and the techniques described in this section have also been used successfully in areas such as inventory control and distribution systems. This section draws on examples from many areas, but will make specific references to fraud risks in accounts payable/purchasing and the analysis of the retail sector using Electronic Point of Sale ("EPOS") data.

The methodology outlined below is based on the use of relatively low cost audit software, and does not necessitate advanced or expensive computer systems. Although, the fraud examiner may well identify that a more sophisticated detection and investigation system is the most appropriate for their organisation. It is a methodology that can be applied to almost any business area where electronically held data exists and where there is an unfulfilled need to analyse and provide exception reports; fraud detection and risk management are just two of these areas.

In practice, the analytical testing process is just one stage of an effective data mining programme. Our experience has shown that effective data mining procedures encompass a number of key elements:

firstly, it is important to lay down the foundations for data mining reviews. For example, ensuring that data matching techniques are, under appropriate registrations, allowable in the jurisdiction. While these techniques may be legal in the UK they may be illegal in other countries and it is vital to review the purpose, ethics and legality of data capture and analysis on the impact of relevant jurisdictional Data Protection Act / privacy legislation or corporate culture;
secondly, the effectiveness of the initial data capture into an electronic form (for example, the input of transactions on to the accounting ledger);
thirdly, the consolidation of data into central data files or data warehouses;
fourthly, the ability to download or transfer the data to an effective and suitable medium (ranging from the use of diskettes to SCSI drives and direct download to the analyst's PC hard drive);
fifthly, the ability to import the data in a usable format into the analysis package; and
finally, the actual testing, review and investigation of findings.

Nowadays companies spend vast amounts of time and effort in capturing data about their businesses. For example, information about clients, suppliers, employees, projects, invoices and budgets is routinely recorded. Unfortunately, all too often this information is held in many disparate systems and is difficult to consolidate. Ad hoc reporting is frequently impossible without diverting specific programming resources from other projects. Such an environment makes automated fraud detection impractical and not cost effective.