Monitoring and detection of NEFT fraudulent requests: A comparative analysis of machine learning models

The world is going through a fast pace of “digital disruption” in business driven by “technology transformation.” Similar to the global trend, the Indian banking industry is experiencing a major digital explosion, especially in the post-demonetization era, and accelerated further with the COVID19 pandemic. Digital channel services create massive volumes of electronic transactions and are automatically processed from end to end. With the increasing use of digital banking and online payments, cyberattacks, money laundering, and other fraudulent activities are rising in frequency and intensity. Frauds are dynamic, show irregular patterns, and are difficult to identify accurately. Cyberattackers use innovative technologies to find the existing systems’ vulnerabilities and siphon millions of rupees. This means traditional fraud monitoring and detection methods are inadequate to meet the challenges that banks and financial institutions face today. This study aims to analyze the results of applying Machine Learning (ML) algorithms to digital payment transactions and measure the accuracy of segregating valid and abnormal transactions. The methods used are Logistic Regression (LR), k-Nearest Neighbor (KNN), Support Vector Machine (SVM), Naïve Bayes (NB), and Decision Tree (DT). The datasets used are the National Electronic Fund Transfer (NEFT) N02 messages. We analyze five evaluation metrics - Precision, F-Score, Accuracy, Misclassification Error, and Area Under the Receiver Operating Characteristic Curve (AUC). The results propose the models in online payment systems for continuous monitoring and early detection of possible fraudulent transactions.