Tóm tắt Luận văn Thạc sĩ: Nghiên cứu phát hiện gian lận trong giao dịch thẻ tín dụng dựa vào Học sâu

The rapid expansion of digital banking and smart device usage has revolutionized consumer behavior, leading to a surge in online transactions. This digital transformation, however, has also presented new vulnerabilities, making financial fraud, particularly credit card fraud, a growing global concern. The increasing volume of transactions and the demand for faster processing necessitate sophisticated fraud detection and prevention mechanisms. Traditional rule-based systems often struggle against the dynamic and evolving nature of fraudulent activities, where perpetrators constantly adapt their methods to evade detection. This research aims to address these critical challenges by leveraging deep learning, specifically Autoencoder models, to enhance the accuracy and efficiency of fraud detection in credit card transactions, thereby safeguarding financial systems and consumer trust.

Chuyên gia công nghệ thông tin, nhà nghiên cứu học máy, sinh viên và giảng viên ngành hệ thống thông tin và an toàn tài chính, cùng các chuyên gia trong lĩnh vực ngân hàng và phát hiện gian lận.

This Master's thesis delves into the critical issue of credit card fraud detection, a growing concern amid the expansion of digital banking. It highlights the limitations of traditional rule-based and supervised machine learning approaches in combating the ever-evolving tactics of financial fraudsters, who constantly change their methods to avoid detection and exploit new technologies. The research posits that unsupervised learning offers a more robust solution, particularly for identifying novel and unforeseen fraud patterns. The core methodology involves the application and evaluation of Autoencoder models, a type of Artificial Neural Network, for anomaly detection. The study emphasizes how Autoencoders are trained to effectively reconstruct normal transaction patterns; any significant deviation or reconstruction error signals a potential fraudulent activity. This approach is particularly effective for imbalanced datasets, common in fraud detection, where genuine transactions vastly outnumber fraudulent ones. The research utilizes an experimental dataset from Kaggle to build, train, and assess the Autoencoder model's performance in real-world credit card transaction scenarios. The theoretical background covers various fraud detection techniques, including supervised, semi-supervised, and unsupervised learning, along with detailed explanations of neural network architectures. The findings of this research contribute significantly to the field of financial security by providing an advanced, AI-driven framework for combating credit card fraud. By leveraging deep learning, specifically Autoencoders, financial institutions can enhance their ability to detect subtle and emerging fraudulent activities with greater accuracy and efficiency, thereby minimizing financial losses and bolstering consumer trust. This work has practical implications for banks and payment service providers seeking to upgrade their fraud detection systems to be more adaptive and intelligent against sophisticated cyber threats.