An Evaluation of Machine Learning Techniques for Detecting Banking Frauds

Abstract

Credit cards are an essential part of every person's daily life in the modern world. Customers use their credit cards and online transactions to make purchases for their requirements. To reduce the risk of defaulters, banks and other financial institutions consider customer credit applications. Credit risk is the increase in debt owed by a customer who repeatedly misses payments on their bills. The precision and recall measures received greater attention than the metrics for accuracy. Based on the False Negative value of the confusion metrics, logistic regression is the best model after comparison with the precisionrecall curve. Additionally, a GUI (Graphical User Interface) was built after altering the logistic regression's threshold value and it successfully forecasted whether a customer was a defaulter or not. It is a problem to categorize clients' credit payment defaulters and nondefaulters. The basic objective of risk prediction is to evaluate each customer's credit risk and reduce it using information from financial statements, customer transactions, and repayment history. The damage and unpredictable nature. Different methods, including logistic regression and decision trees, are used to create risk prediction models. © 2023 IEEE.