Four Fraud Detection Research Papers Worth Reading

When most people hear the word Ekata, they think of phone numbers and phone books. But Ekata is far more than a phone number and address directory. Ekata has become a leader in global identity verification data, thanks to products like Identity Check featuring the Confidence Score – powered by a team of data scientists and product managers that live and breathe machine learning. So we thought we would share a few research papers that cover a variety of machine learning techniques for fraud detection that we’ve found interesting. All of these papers are available for download at arXiv.org.
A Survey of Credit Card Fraud Detection Techniques: Data and Technique Oriented Perspective
This research paper by Sorournejad, Zojaji, Atani, & Monadjemi (2016) provides the results of an investigation regarding the difficulties of credit card fraud detection. The paper describes how many fraud detection systems are prone to a number of challenges which includes overlapping data (false positives and false negatives) and lack of adaptability. The paper goes into detail about the types of credit card fraud detection techniques the researchers tested. Among the fraud detection methods tested are artificial neural network (ANN), Bayesian network, and decision tree (DT). The paper features several tables including one that describes the disadvantages and advantages of each fraud detection method and another that provides details of the different data sets used.
Spectrum-Based Deep Neural Networks for Fraud Detection
Another interesting research paper about fraud detection is by Yuan, Li, Wu, & Lu (2017). This research paper focuses on fraud related to online social networks (OSNs), specifically identity fraud, fake content, and fraudulent activities. The researchers present a framework that combines deep neural networks and spectral graph analysis. The first step of the framework involves obtaining node spectral coordinates by conducting graph spectral projections on a signed graph. Then the researchers used a combination of the node and its s-step neighbors’ spectral coordinates as inputs for the fraud detection models (convolutional neural network) and deep autoencoder.
Solving the “False Positives” Problem in Fraud Prediction
False positives are a common problem in the fraud detection and prevention industry. This research paper by Wedge, Kanter, Veeramachaneni, Rubio, & Perez (2017) focuses on reducing false positives in fraud prediction with an automated feature engineering-based approach. The researchers used the Deep Feature Synthesis (DFS) algorithm from Featuretools, an open source framework for automated feature engineering, to automatically generate behavioral features from the historical data of the credit card associated with a transaction. The paper states that the machine learning model was tested on data from a large multinational bank. The solution was able to reduce the number of false positives by 54% compared to the bank’s existing fraud prediction solution. The test size was unseen data of 1.852 million transactions.
Transaction Fraud Detection Using GRU-Centered Sandwich-Structured Model
This research paper by Xurui Li et al. (2017) presents a sophisticated four-step solution for detecting fraud. The method is described in the paper as a new “within->between->within sandwich-structured sequence learning architecture.” The first step involves using Spark to perform feature engineering. The second step involves optimizing features within a single transaction using a gradient boosted decision tree (GBDT) model. The next step involves improving how the model learns relationships between transactions by applying a gated recurrent unit (GRU) model on transformed sequential samples. The final step uses optimized transaction eigenvectors to train a top-layer RF classifier.
These are just a few of the really interesting research papers available on arXiv about using machine learning techniques for fraud detection and prevention.
If you are interested in learning more about our approach to machine learning and how it feeds our solutions, reach out to one of our data experts.

Start a Free Trial

See how Ekata can reduce fraud risk for your business, contact us for a Demo.