Credit Card Fraud Detection Using Machine Learning Based on Support Vector Machine

• Published in: 2023 Eighth International Conference on Science Technology Engineering and Mathematics (ICONSTEM) pp. 1 - 6
• Main Authors: Priyaradhikadevi, T., Vanakovarayan, S., Praveena, E., Mathavan, V., Prasanna, S., Madhan, K.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 06-04-2023
• Subjects: CCF; Credit cards; deep learning; Fraud detection; Machine learning; Machine learning algorithms; Quantum annealing; Real-time systems; Support vector machines; Time series analysis
• DOI: 10.1109/ICONSTEM56934.2023.10142247

The usage of machine learning algorithm in the detection of fraudulent transactions is becoming more common. Most application systems, on the other hand, only catch fraudulent activityafter it has already taken place, rather than in real time. Detecting fraud is difficult because there are considerably fewererroneous transactions than legitimate ones. This data imbalance necessitates methods other than machine learning to handle it. Quantum machine learning (QML) has been used to develop detection frame work, which was then, implemented using SVMs supplemented with quantum annealing solvers. A total of twelve machine learning algorithms have been applied to testQML's detection capability, and their results have been compared to those of the QML application ontwo datasets: a non-time series of Israeli credit card transactions and a time series of Israeli bank loanapplications. The results reveal that, using the bank loan dataset, the quantum augmented SVM over takes the others in relationships of both speed and accuracy. The detection accuracy is comparable to those that use Israel credit card transaction data.By the detection time can be greatly improved for both data sets by using feature selection, although the increase in accuracy is minor. QML applications on time series data with significant imbalance have been shown to have promise, whereas standard machine learning methodologies have been shown to have worth when dealing withnon-time series data, as these results show. This research sheds light on how to choose the besttechnique for various datasets while keeping in mind the trade-offs between speed, accuracy, and price.