Data Driven Analysis of Fraudulent Transaction Characteristics in Branchless Banking

In branchless banking, fraudulent transactions can be defined as an act of an agent to conduct some non-essential transactions to receive transaction's fee. This transaction is a legitimate transaction, in terms of money movement. It is hard to detect the fraudulent transaction in this manner. This study aims to learn about the characteristics of a fraudulent transaction and provide a rule for determining suspected fraudulent transaction. To reduce the size of transactions data, Leiden community detection algorithms is used. The user is clustered in a transactional graph data. Each community is then statistically processed to acquire the outlier. From the analysis it is estimated that 25% of agent has done fraudulent transactions. The fraudulent transaction characteristic was expressed in term of the average of the community. The fraudulent transactions are 185% above the average in terms of transactions value. While for the transaction frequency, the fraudulent transactions are 90% above the average frequency. The fraudulent transactions data always act as an outlier in their respective community.