Fraud Detection in Financial Transactions

Please enable javascript in your browser.

Haghighat, Mohammad | 2023

87 Viewed

Type of Document: M.Sc. Thesis
Language: Farsi
Document No: 55941 (19)
University: Sharif University of Technology
Department: Computer Engineering
Advisor(s): Beigy, Hamid
Abstract:
With development of electronic payment infrastructures and increase of payment transactions in result, abusing these infrastructures and fraudulent efforts has been increased. Problem of “Fraud Detection in Financial Transactions” is finding these illegal/abnormal transactions while many other legitimate transactions exist. Goal of this thesis is providing a method for fraud detection in financial transactions using representation learning. Many approaches are used for solving fraud detection including classic data mining algorithms and deep learning based methods, which are compared in this thesis. We also covered diverse feature engineering and representation learning ideas for improving previous works. Afterward we tried to study impact of most significant ideas on quality of fraud detection. In this thesis we improved SubTab framework at facing tabular data, such as payment transactions. These improvements caused 18% increase on AUC ROC of famous IEEE-CIS Fraud Detection in compare with original SubTab framework. At the end we mentioned some further works for using representation learning on problems related to tabular data
Keywords:
Representation Learning ; Financial Fraud Detection (FFD) ; Machine Learning ; Neural Network ; Data Mining ; Tabular Data

No TOC