Enhanced stacking classifiers system for credit card fraud detection

Credit cards have been an option for payment method by many people nowadays. Despite that, without handling the usage of credit cards carefully, it may lead to undesirable incidents such as credit card fraud activities. Credit card fraud involves the illegal use of a card or card information without...

Full description

Saved in:
Bibliographic Details
Main Author: Ishak, Nur Amirah
Format: Thesis
Published: 2022
Subjects:
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Credit cards have been an option for payment method by many people nowadays. Despite that, without handling the usage of credit cards carefully, it may lead to undesirable incidents such as credit card fraud activities. Credit card fraud involves the illegal use of a card or card information without the owner’s knowledge. Credit card fraud was estimated to exceed a $35.5 billion loss globally in 2020. It has given direct or indirect financial loss to its holders. Due to this, a detection system capable of analysing and identifying fraudulent behaviour in credit card activities becomes essential to develop. However, credit card data is not easy to handle because of its characteristics, such as (i) unbalanced class distributions and (ii) overlapping of classes that causes difficulties for general classifiers to detect fraud, especially the minority. Such characteristics challenge researchers to find a seamless approach to address minority fraud using data mining algorithms. Considering the characteristics of credit card data sets and the weaknesses of general classifiers, this project proposed to detect credit card fraud using Enhanced Stacking Classifiers System (ESCS). ESCS is a multiple classifiers system utilising a hybrid scheme. It comprises two sequential levels, and it has been designed by separating the classes and tackling the data individually. The first level contains a classifier that is strong in detecting normal credit card transactions (the majority class). The second level contains a single-level stacking classifier that is good in distinguishing credit card frauds (the minority class). The ESCS can improve the fraud detection rate via the second level as it contains sensitive classifiers to learn and identify the misclassified fraud transactions as normal transactions from the first level. The meta-classifier then combines the decisions of all the base classifiers from the levels to produce the final detections. Three experiments, namely, single classifiers, ensemble classifiers, and ESCS were conducted in this project on two data sets that comprise the two main characteristics mentioned earlier. They are Credit Card Fraud Data Set (CCFD) and UCSD-FICO Data Mining Contest 2009 data set. The performance of ESCS outperforms the others. For CCFD, the highest TPR for the minority class (frauds) is 0.8841 using ESCS. For UCSD-FICO Data Mining Contest 2009 data set, the highest TPR for the minority class (frauds) is 0.5203 using ESCS. On top of that, ESCS also outperforms existing researcher works, especially in detecting fraud of these two credit card data sets. This proejct proves that ESCS can improve the fraud detection rate on credit card data.