Cost-Sensitive Model Evaluation Approach for Financial Fraud Detection System

Global statistics shows financial institutions lose millions of dollars due to transaction frauds every year. Fraud detection involving manual observations becomes impossible with millions of electronic transactions (e-tail) being done every minute. Fraud detection has been defined as a problem with huge scope of research, many academic researchers have designed and developed solutions for financial fraud detection but most of the work and their application gets scrapped by the financial institutions. The major reason observed and highlighted by many industry experts has been the lack of understanding about the working of the fraud systems and the cost calculation attached to each transaction. This article discusses the various machine learning algorithm used for fraud detection by the financial institutions along with the data processing techniques, and aims at providing an insight on the working of a financial fraud protection system, the process of raising a fraud flag, the cost components attached to each transaction and the impact of each misclassification made by the machine learning model.