CREDIT CARD FRAUD DETECTION WITH AUTOENCODERS, ONE-CLASS SVMS AND ISOLATION FORESTS

Date

2023-12-6

Author

Özkum, Özdemir

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

293
views

638
downloads

Cite This

Fraud Detection has gained importance in the last few years due to its cost to the economy and the population in general. However, there are a number of important problems in the modeling process to tackle when detecting fraud. One is the availability of data. Since the financial data produced by the customers is subject to privacy rules, working with synthetic data is a necessity. While this is the case, there are very few data sets that reflect the true nature of the data generation and analysis processes of financial transactions. Another problem is that the fraud or any data that require anomaly detection do not have labeled data. Even labeled, the data points are labeled by rules and there is a need to automate this process. This is why it is deemed important to incorporate an unsupervised collection of methods and let the data speak for itself. For these reasons, this study focuses on the unsupervised learning methods in fraud detection using a synthetic data set which consists of credit card transactions. The learning methods used were the Autoencoders, Sparse Autoencoders, One-Class Support Vector Machines and Isolation Forests. For comparison, a Random Forest Model was also built. It was found that among the unsupervised methods One-Class SVM was the best performing model with Precision=0.68, Recall=0.98, and F1 Score=0.81. However, One-Class SVM did not outperform the supervised Random Forest model which achieved Precision=1.00, Recall=0.95, and F1-Score=0.97.

Subject Keywords

Denetimsiz Öğrenme, Denetimli Öğrenme, Kredi Kartı Dolandırıcılığı, Otokodlayıcılar, Bir Sınıflı Destek Vektör Makineleri, İzolasyon Ormanları, Rassal Ormanlar

URI

https://hdl.handle.net/11511/107729

Collections

Graduate School of Natural and Applied Sciences, Thesis