Development of a banking risk-engine to predict credit, market and operational risks using machine learning

Abstract

Effective management of credit risk is an essential component of any banking system. The identification of credit card fraud in financial transactions is one of the most pressing issues facing financial institutions today. Fraud involving credit cards is significantly on the rise as modern technology continues to advance on a daily basis. Credit card fraud results in annual losses of billions of dollars for a number of different financial institutions. So, the utilization of fraud detection strategies is essential for both banking and non-banking financial institutions in order to reduce the amount of money lost. In order to anticipate credit card fraud detection under credit risks of banks, the use of machine learning (ML) methods can be effective. Appropriate ML algorithms have the potential to identify fraudulent and lawful transactions. However,there is difficulty in exchanging datasetsand ideas of fraud detection among different banks due to privacy issues, and the lack of datasets is an obstacle tocredit card fraud detection techniques. This work focuses mostly on analyzing the effectiveness of multiple ML classifiers, such as Random Forest (RF), AdaBoost, and CatBoost, with the intention of categorizing fraudulent behaviors involving credit cards.The dataset considered in this research is the transactions performed with credit cards by European cardholders in September 2013 where there are 492 fraudulent transactions, and the total number of transactions is 284,807. Results show that RF and CatBoost achieve an accuracy value of 99.92%, while AdaBoost exhibits an accuracy value of 99.91% for fraud detection.In the future, there will be a need for large real-time datasets to train the model while protecting privacy