Machine Learning Algorithms for Anomalous OTP Usage Pattern Detection: A Comprehensive Review

Abstract

One-Time Passwords (OTPs) serve as an additional level of security for online authentication and digital transactions. While generally effective, OTP systems have become an increasingly popular target for fraudsters using phishing, SIM swapping, and social engineering. These risks point to the need to detect unusual or suspicious patterns of OTP use before attempts lead to fraud. This study evaluates the use of machine learning algorithms to tackle this challenge. We go through different manoeuvres, applying supervised models like decision trees and support vector machines to discuss unsupervised methods like clustering and anomaly detection techniques. The review highlights important issues such as imbalanced data, high false positive rates, and the inability to respond to new fraud strategies, while also looking into more recent developments like ensemble, graph-based, and federated learning as intriguing areas of development. Brokerage points towards bringing all this together, we evidence a clearer overview of the state of research and ways in which we might practically go about developing higher quality, smarter, and adaptive OTP fraud detection systems.