Aristotle University Of Thessaloniki (AUTH)
Εξόρυξη χρηματοοικονομικών δεδομένων με εφαρμογή στη λογιστική
Abstract
dc:descriptionDuring the last years, Data Mining provides new methodologies for the extraction of knowledge from large amounts of data. Moreover, classification methods like Neural Networks and Decision Trees allow the development of predicting models. The need to employ these methodologies to cope with problems related to Accounting and Au-diting, has been highlighted by numerous researchers and relevant professional and scientific organizations. The present thesis comes to contribute to this direction by de-veloping predicting models and revealing significant factors, related to specific tasks in the field of Auditing.The first research subject was the detection of fraudulent financial statements. These financial statements have been falsified by the management to deceive inves-tors, creditors or the tax authorities. Three classification methods, i.e. Decision Trees, Neural Networks and Bayesian Belief Networks have been applied. The data used were about Greek companies. The Bayesian Belief Network model achieved the best performance in predicting the class of unknown observations. The interpretation of the models reveals that financial distress and profitability are significant factors, which are strongly associated with the falsification of the financial statements.The second research subject was the prediction of the cases for which the exter-nal auditors issue a qualified opinion during the audit of the annual reports. For this research study, data concerning British and Irish companies have been used. Three alternative predicting models were developed and compared. The employed classifica-tion methods were Decision Trees, Neural Networks and Bayesian Belief Networks. According to the results, the Bayesian Belief Network model outperformed the other two models in predicting the class label of unknown observations. All three models agree that financial distress and profitability are significant factors, influencing the auditors in the issuance of a qualified opinion.The third research subject was the prediction of the type of external auditor that companies select. The type of the external auditor is indicative of the level of audit quality that a company aims at. In this research study, data about British and Irish companies were used. Three classification methods, i.e. Decision Trees, Neural Net-works and k-Nearest Neighbor were employed to predict the cases were companies hire a big international auditing firm. The Decision Tree model achieved the best per-formance in predicting the type of auditor of unknown observations. According to the results, the debt level is the most significant factors in the auditor selection decision. Τo increase the models’ performances we applied bagging to the three classifiers. Bagging improved significantly the accuracy rate of the Decision Tree model.The findings of the conducted research can be useful to internal and external auditors, financial analysts, credit risk estimators and tax authorities. This research also offers considerable contribution to the development of sophisticated tools for audi-tors.
Degree
thesis:*- Grantor dc:publisher
- Aristotle University Of Thessaloniki (AUTH)
- Year dc:date
- 2007
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Kirkos, Efstathios
- Κύρκος, Ευστάθιος
Subjects
dc:subject × 12- Εξόρυξη χρηματοοικονομικών δεδομένων
- Νευρωνικά δίκτυα
- Δένδρα αποφάσεων
- Μπαϋεσιανά δίκτυα πίστης
- Κατηγοριοποιητές βασισμένοι σε παραδείγματα
- Εντοπισμός παραποιημένων χρηματοοικονομικών καταστάσεων
- Πρόβλεψη σχολίων εξωτερικών ελεγκτών στις οικονομικές καταστάσεις
- Πρόβλεψη του τύπου του εξωτερικού ελεγκτή
- Φυσικές Επιστήμες
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Natural Sciences
- Computer and Information Sciences
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/30247