Credit Card Fraud Detection using Pipeling and Ensemble Learning

TLDR

Financial fraud poses a significant threat to the financial industry, and credit card fraud detection is particularly difficult due to evolving fraud patterns and highly imbalanced datasets. The study aims to compare the performance of multiple classification algorithms on credit card fraud data. The authors evaluate logistic regression, K‑nearest neighbors, random forest, naive Bayes, multilayer perceptron, AdaBoost, quadratic discriminant analysis, pipelining, and ensemble learning.

Abstract

Financial fraud is a problem that has proved to be a menace and has a huge impact on the financial industry. Data mining is one of the techniques which has played an important role in credit card fraud detection in transactions which are online. Credit card fraud detection has proved to be a challenge mainly due to the 2 problems that it poses - both the profiles of fraudulent and normal behaviours change and data sets used are highly skewed. The performance of fraud detection is affected by the variables used and the technique used to detect fraud. This paper compares the performance of logistic regression, K-nearest neighbors, random forest, naive bayes, multilayer perceptron, ada boost, quadrant discriminative analysis, pipelining and ensemble learning on the credit card fraud data.