Date of Award


Document Type


Degree Name

Master of Science in Information Systems and Technology


Information and Decision Sciences

First Reader/Committee Chair

Shayo, Conrad


This Culminating Experience Project explores the use of machine learning algorithms to detect credit card fraud. The research questions are: Q1. What cross-domain techniques developed in other domains can be effectively adapted and applied to mitigate or eliminate credit card fraud, and how do these techniques compare in terms of fraud detection accuracy and efficiency? Q2. To what extent do synthetic data generation methods effectively mitigate the challenges posed by imbalanced datasets in credit card fraud detection, and how do these methods impact classification performance? Q3. To what extent can the combination of transfer learning and innovative data resampling techniques improve the accuracy and efficiency of credit card fraud detection systems when dealing with imbalanced datasets, and what novel strategies can be developed to address this common challenge?

The main findings are: Q1. Unconventional cross-domain methods improved fraud detection, holding promise for enhanced security. Q2. The problems caused by unbalanced datasets in credit card fraud detection were effectively addressed by the synthetic data generation techniques SMOTE and ADASYN, resulting in a more balanced dataset suitable for fraud classification. Q3. The combination of neural networks and data resampling techniques, such as SMOTE and ADASYN, significantly improved credit card fraud detection accuracy.

The main conclusions are: Q1. Cross-domain methods are useful for credit card fraud detection, especially when it comes to online transactions. Q2. When used with various classifiers, neural networks show remarkable accuracy rates: 97% for unbalanced data, 99.47% for SMOTE, and 99.11% for ADASYN Q3. A fraud recall of 0.99 is obtained by the model evaluation on imbalanced data, with 12,155 right predictions out of 12,336 and 181 incorrect ones. The identified areas for further study encompass the testing of our model on larger datasets and the optimization of hyperparameters for further enhancement.