Comparative Analysis of Applications of Machine Learning in Credit Card Fraud Detection

Recent statistics show that Credit card fraud had increased by a hefty 20% in 2021, marking a stark rise in criminal activity from the previous year. Several techniques have been applied so far to counter the rise of fraudulent activity. However, there still lies scope for improvement in the techniques applied. One of the many ways that a fraud detection system can be improved is by the introduction of machine learning into its applications. The key purpose of a card fraud detection program is to identify fraudulent or suspicious activity and report them to an analyst whilst letting the normal transactions pass through. With the application of supervised algorithms, like Logistic regression and K-Nearest Neighbors, the efficiency for detecting fraudulent cases, by systems that have trained on detection of suspicious transactions, becomes all the more efficient. In this paper, we present a comparative analysis detection of fraud cases through usage of supervised algorithms, an AutoEncoder (neural network architecture), and one more model, OCSVM (One-Class Support vector Machine), from a library built specifically for outlier detection. Metrics like precision, recall and F1-score are used to evaluate the models and select the optimal algorithm for credit card fraud detection, which results in a balanced combination of the metrics used to evaluate them. The selected algorithm will help prevent such suspicious activities with regard to credit card transactions, with prompt accuracy and precision.

Перевод пока недоступен