The Challenge of Quality Evaluation in Fraud Detection

A system is considered to be safety-critical when its failure may lead to unacceptable consequences [17]. One failure of particular interest in safety-critical systems is fraud, understood as abusing a position of trust to get illegal advantage and/or cause losses by false suggestions or suppression of the truth. Although some safety-critical systems are protected by design against known threats, innovative exploitation of latent vulnerabilities remains a possibility. Fraud has multiple consequences on provision of services, management of organizations, and operational infrastructure, causing damage to human lives, production processes, and the natural environment. Paradoxically, in fraud situations, compliance to fitness for use seems to be appropriate, while conformity to procedures is cleverly bypassed. Also, even if automatic procedures continuously check for potential fraud, legitimate users may commit fraud inside organizations or maliciously manipulated applications can have significant unknown repercussions on a system. Besides, voluminous heterogeneous data and information are exchanged in and between systems. For these reasons, quality evaluation in fraud situations is very complex. The automobile industry [14], banks [7], health institutions [21], and government [2] are some examples of organizations affected by frauds. Taking into account that the quality assessment of procedures and processes is still an open problem [1] [5], this proposal focuses on data and information quality within those procedures. This article discusses why quality assessment in fraud contexts is a challenging problem and argues for a longitudinal quality meta-analysis, relying on contextual cumulative indicators.