A Novel Multivariate Classification Model using Triangular Boundary based Outlier Detection (TBOD) Algorithm to Discover Suspicious Transaction Activity

There is an immense regulatory overhaul over the Money Laundering (ML) threats in financial systems. To build trust and confidence there is an industry wide push to invest more in the Anti-Money Laundering (AML) programs which includes Know Your Customer (KYC), Transaction Monitoring(TM), Trade Surveillance, Fraud Detection and Prevention. These great deals of critical processes require detecting money laundering in the real world financial systems in response to mounting concern over money laundering. Data mining plays a major role to accommodate various types of meaningful and useful information from banking systems. Outlier detection mechanism helps to discover the abnormal behavior in the dataset to identify the suspicious activities in today's world. This paper proposes a novel technique called Triangular Boundary based Outlier Detection (TBOD) Algorithm to detect suspicious activity in banking environments. Segmentation is the preliminary step applied on the dataset to cluster/group the customer based on product usage and associated customer risk. For each customer segment, the profile is modeled to set the boundaries based on historical transaction behaviors. Then, Multivariate classification is applied on the clustered dataset to identify the outliers. Distinct TBOD algorithm has been proposed to model the system and validate it to detect suspicious transactions. The records are arranged as triangular matrix, which visualizes the upper and lower levels of records. The abnormal behavior is identified based on the boundary condition proposed. Various performance measures have been taken to examine the results. The proposed method results in higher precision, recall, sensitivity and specificity values with lesser MAE than the existing SVM K-cross validation, LDOF and K-Means IBK algorithms. It guarantees the accurate detection of outliers in the dataset.