<![CDATA[The rise in credit card transactions has been accompanied by an increase in fraudulent activities. One of the key challenges in detecting fraud is the distribution of the dataset, where fraudulent transactions are significantly outnumbered by normal ones. Despite their low occurrence, fraudulent transactions have a significant impact on the banking sector. Therefore, an effective model is needed to identify and estimate fraudulent transactions. This study aims to generate optimal training dataset from an imbalanced one using Adaptive Synthetic Sampling (ADASYN) to enhance the training process of Support Vector Machine (SVM) model. The dataset used consists of anonymized credit card transactions and labeled as either fraudulent or normal, sourced from the Kaggle dataset. It contains transactions made by European cardholders in September 2013, covering a two-day period with 492 fraud cases out of 284,807 transactions. Three datasets were derived from the original: raw, balanced, and support vector-based balanced. The SVM model training on these datasets resulted in sensitivities of 0.39, 0.64, and 0.70, respectively, while the precision values were 0.92, 0.72, and 0.01. The corresponding f-measure values were 0.55, 0.68, and 0.02. The best performance based on the f-measure was achieved using the balanced version of the raw dataset.]]>
Copyrights © 2025
