Credit Card Anomaly Transaction Detection Using Adaptive Synthetic Resampling (ADASYN) and Support Vector Machine (SVM)

This title has been presented on Friday, December 15, 2023 at 15.50-16.00 GMT+7.

Abstract

This title has been presented on Friday, December 15, 2023 at 15.50-16.00 GMT+7.

There are anomalies or fraudulent transactions due to the large number of transactions that occur on credit cards, making it necessary to detect anomalies in credit card transactions so as not to harm customers or banks. One of the obstacles in detecting these anomalies is the formation of mathematical models, namely, datasets that have an unbalanced class distribution. One example is anomalous credit card transactions, which are fewer in number than in normal transaction data. Although transactions are low in number, they have a significant impact. Therefore, a model is required to perform screening to estimate fraudulent transactions. The purpose of this study is to produce ideal training data from an unbalanced dataset using Adaptive Synthetic Sampling (ADASYN) so that it can be used for the Support Vector Machine (SVM) model training process. The dataset is anonymized credit card transactions labeled as fraudulent or genuine, which is available on the Kaggle dataset. The dataset contains transactions made by European cardholders using credit cards in September 2013. This dataset presents transactions that occurred in two days, of which 492 were frauds out of 284,807 transactions. Three datasets were created based on the main dataset: raw, balanced, and balanced support vector. SVM model training using the datasets gain sensitivity 0.82, 0.98, and 1. The specificity are 0.99, 0.99, and 0.74, respectively. Model training using balanced support vector data can detect all anomalous transactions.