Digital Fraud and Cybersecurity: Artificial Intelligence-Based Strategies for Anomaly Detection

Abstract

Fraud in electronic transactions represents a growing challenge in the financial sector, driven by digitalization and the rise of online commerce. Cybercriminals have developed increasingly sophisticated strategies to exploit vulnerabilities in payment systems, resulting in significant financial losses and compromising the security of consumers, businesses, and banking institutions. Traditionally, fraud detection has relied on predefined rules and supervised models, which require large volumes of labeled data. However, the rapid evolution of fraudulent tactics limits the effectiveness of these approaches. In this context, anomaly detection-based machine learning emerges as an alternative for the early identification of suspicious transactions without the need for prior fraud data. This study focuses on developing a model based on anomaly detection techniques to identify real-time fraudulent transactions. Various approaches will be evaluated, including autoencoders, isolation forests, and One-Class SVM. Autoencoders, as neural networks designed to reconstruct normal data, can detect suspicious transactions when reconstruction errors are high. Isolation forests identify anomalies by isolating outlier observations in a dataset, enabling efficient fraud detection. Finally, One-Class SVM creates a decision boundary that separates normal transactions from potentially fraudulent ones, making it particularly useful in scenarios where fraud cases represent a small proportion of total transactions. The implementation of these techniques will allow the analysis of large volumes of data with greater accuracy and speed, facilitating more effective fraud pattern detection. The results obtained will contribute to the development of more efficient solutions for protecting electronic transactions in financial and commercial environments.