Fraudulent credit card transactions are on the rise and have become a significantly problematic issue for financial intuitions and individuals. Various methods have already been implemented to handle the issue, but the embezzlers have always managed to employ innovative tactics to circumvent a number of security measures and execute the fraudulent transactions. Thus, instead of a rule-based system, an intelligent and adaptable machine learning based algorithm should be an answer to tackle such sophisticated digital theft. The presented framework uses k-NN for classification and utilises Principal Component Analysis (PCA) for raw data transformation. Neighbours (anomalies in data) were created using Synthetic Minority Oversampling Technique (SMOTE) and a distance-based feature selection method was employed. The proposed process performed well by having a precision and F-Score of 98.32% and 97.44% respectively for k-NN and 100% and 98.24% respectively for Time subset when using the misclassified instances. This work also demonstrates a larger and clearer classification breakdown, which aids in achieving higher precision rate and improved recall rate. In a view to accomplish such high accuracy, the original datum was transformed using Principal Component Analysis (PCA), neighbours (anomalies in data) were created using Synthetic Minority Oversampling Technique (SMOTE) and a distance based feature selection method was employed. The proposed process performed well when using the misclassified instances in the test dataset used in the previous work, while demonstrating a larger and clearer classification breakdown.
|Number of pages||27|
|Journal||International Journal of Electronic Commerce Studies|
|Publication status||Published - 2019|