The discovery of the credit card transactions suspicious of fraud using unsupervised data-mining methods (single-link hierarchical clustering)

Zohreh Darbandian, Alimohammad Latif, Sima Emadi

Abstract


Fraud, the intentional misuse of the resources for personal interest, is an unlawful act. The discovery of the fraudulent actions is an operation which takes place in respect to concrete decision making regarding a suspicious behavior. The method proposed in the present dissertation is the discovery of fraudulent acts by making use of the unsupervised data mining methods (single-linkage hierarchical clustering method) by the application of which the amount of the data suspicious of cheating can be identified in credit cards. The collection of the data gathered for this purpose pertained to the transactions carried out by some individuals in 2013, and the data were extracted from a state bank in Tehran. The data analysis and methods implementation stages have been conducted by taking advantage of R-software. The (unsupervised) data-mining technique provides for better results in respect to the other algorithmic data-extraction methods. Due to the confidential nature of the bank data in Iran, the available data for the identification of the data suspicious of being fraudulent are unlabeled (fraudulent and non-fraudulent together) data. The results obtained from the present study indicated meaningful and sensible findings and the method used by the current thesis article was applied to reveal the suspicious cases from among the great many of the performed transactions and then in the end it enabled the bank supervisors to only scrutinize the suspicious cases with a higher precision and it was not necessary to survey the entire volume of the bank transactions which were too many.


Keywords


fraud, data-mining, clustering, suspicious bank data.

Full Text:

PDF

Refbacks

  • There are currently no refbacks.