Guarding the Digital Checkout
Open Access DepositedLeveraging Data-Driven Strategies to Detect and Prevent Fraudulent E-Commerce Transactions
In the realm of cybersecurity, there are many different concerns in the world today and there are surely many more that will arise. One substantial issue that impacts businesses and consumers throughout different industries is fraudulent activities. As a part of purchasing activities, many users partake in online shopping. Online shopping makes up a large portion of fraudulent experiences. When conducting transactions online, there is always a minor level of risk, because the identity of the party making the purchase is not truly known. There is an existing authentication protocol, 3-D Secure, that was developed to provide more confidence when a user is checking out online. Even with this in place, fraudulent transactions still make their way through, causing financial losses, compliance concerns, potential reputational harm, and more. This praxis aimed to add an extra layer of defense to the already existing authentication transaction flow, by building a hybrid machine learning model to use in tandem. The hybrid machine learning model leveraged the unsupervised anomaly detector Isolation Forest technology with the supervised classifier Neural Network technology. Feature engineering was performed on the original feature set to derive patterns that the models may not have otherwise picked up on. The Isolation Forest produced anomaly score like features that were appended to the existing feature set and fed into the Neural Network model. Due to the nature of fraudulent transaction data having a severe class imbalance, as that is how it is realistically presented in the real world, Synthetic Minority Oversampling Technique was applied to the training data only. This allowed the Neural Network to have more fraudulent samples to learn patterns from. The models were evaluated on several validation metrics. This was all completed while keeping end-to-end processing time in mind. In order for this type of model to be used in real-time processing, to assist in detecting and preventing fraud, the latency added needed to be as minimal as possible.
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.
| Thumbnail | Title | Date Uploaded | Visibility | Actions |
|---|---|---|---|---|
|
|
AlHajabed_gwu_0075A_17669.pdf | 2026-02-26 | Open Access |
|