Using precedents to identify top management fraud : the study of a case-based learning and reasoning model.
This thesis discusses how best to design, implement and evaluate a Case-Based Learning and
Reasoning (CB-LR) model to assist accountants in identifying Top Management Fraud (TMF).
There is no structured knowledge about TMF in the fonn of rules, only cases encountered by
experienced auditors. The changing economic, financial and social environment has produced
more fraud which at the same time has become increasingly complex to identify and isolate.
Previous research shows that fraud has evaded auditors, and highlights a need for new
computer-based learning and reasoning paradigms in this domain.
Case-Based Reasoning (CBR) has been considered as an approach to building knowledge
systems which involves reasoning about the current situation on the basis of pinpointing and
resuscitating past instances. As with artificial intelligence in general, there is no standard readymade
CBR method suitable for any domain or application: the challenge in CBR is to come up
with methods that are suitable for problem-solving and learning in particular subject domains
and for particular application environments. INCASE, a Top Management Fraud diagnostic
application, has been designed with a specific methodology derived from Schank and
Riesbeck's principles. INCASE works on the basis of the interviewees' concerns so that it can
act as a 'stimulus agent' for decision support. The intention is for auditors to use it proactively
in a conversational mode, focusing on learning and reasoning about the problem domain~ hence
the use of the CB-LR model.
Although evaluation methods relevant to traditional rule-based expert systems have been
discussed in the literature, their emphasis on system issues was found to be inappropriate for
this CB-LR application. Since there is no agreed and established method for evaluating a casebased
tool, 'a new approach is discussed including verification (where the focus is on response
accuracy of infonnation retrieval) and validation (where user judgement and satisfaction are
key issues). One important component in the evaluation was an in-house questionnaire based
on total quality management ideas. Findings from the evaluation of the TMF diagnostic system
suggest that case-based learning and reasoning has a valuable part to play in assisting auditing
profession in the detection of fraud