Year: 2024
Type: article
Abstract: Increasing usage of electronic payment systems has made card fraud a huge concern for cardholders and financial institutions alike.Most traditional methods of fraud detection thrive to maximize the accuracy while being insensitive to the differential cost of false positives -the good transactions wrongly flagged -and false negatives -the frauds not detected.It advocates a costsensitive approach to the detection of card fraud integrating Dynamic Random Forest DRF and K -Nearest neighbors algorithms in order to improve fraud detection accuracy by avoiding cost for misclassifications.One of the major challenges in effective fraud detection is the extraction of relevant features from transaction data.Cardholder spending behaviors do change over time, making more recent transactions more predictive of typical activity for a cardholder.The approach proposed here assigns greater weight to the more recent transactions, letting the model adapt to changed behavior patterns.The DRF algorithm is used in classifying transactions as genuine or forged, which grows with every transaction data, whereas the KNN algorithm supports the finding of local patterns by considering similarities that exist between two transactions.The combination of the two algorithms enables the system to pick up the global and the local spending patterns, and this reinforces the overall process of detecting.The proposed cost-sensitive approach aims at achieving cost-effective misclassifications that balance detection accuracy against the cost.This helps deliver an efficient fraud-detecting system that enhances decision making through better protection of financial consumers as well as financial institutions against fraud.
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W4406039492