A hybrid semi-supervised approach for financial fraud detection

In this paper, we create a semi-supervised methodology for financial fraud detection in bank wire transactions based on a clustering-based-isolation-forest (CBiForest) algorithm. To test this hybrid model, we experiment on wire transaction data of twelve months from China Everbright Bank. The result of abnormal users is proved to be reliable and outperforms other clustering algorithms. Furthermore, our model can be regarded as a huge improvement for traditional expert system in bank.