An examination of machine learning-based credit card fraud detection systems

Himanshu Sinha *

Kelley School of Business Indiana University, Bloomington, Naperville IL Unites State.
 
Research Article
International Journal of Science and Research Archive, 2024, 12(02), 2282–2294.
Article DOI: 10.30574/ijsra.2024.12.2.1456
Publication history: 
Received on 07 July 2024; revised on 15 August 2024; accepted on 17 August 2024
 
Abstract: 
As a result of the e-commerce industry's explosive growth, credit cards are now frequently used for online purchases. In recent years, banks have faced a significant issue with credit card fraud (CCF) due to the difficulty in detecting fraudulent activity within the credit card system. Machine learning is the solution to the problem of CCFD during transactions. An analysis starts with a study of the Kaggle-provided CCFD dataset. There is a considerable disparity between the classifications in the dataset, which has 284,807 transactions total, of which only 492 are deemed fraudulent. The preprocessing methods are used to prepare the data, which includes the handling of the missing values, the detection of outliers, and the encoding of the categorical variables. Five different classification models are tested and evaluated employing different metrics like precision, F1-score, accuracy, and recall. These models are SVM, Random Forest (RF), Bagging, XGBoost, and DT. In terms of spotting fraudulent transactions, XGBoost is the model that has the highest accuracy rate of 99% among the others. To further strengthen the effectiveness and reliability of fraud detection systems, future research might investigate ensemble methodologies and integrate real-time data streams, guaranteeing thorough defence against financial crime.
 
Keywords: 
Credit Card Fraud Detection (CCFD); Security; Financial Fraud; Dataset; Machine Learning.
 
Full text article in PDF: