ARCHIVES

Abstract: The escalating prevalence of financial fraud within the financial sector poses profound challenges. Detecting credit card fraud in online transactions necessitates data mining due to inherent complexities. Addressing two key issues—evolving patterns in legitimate and fraudulent behaviors and highly skewed datasets of credit card frauds—renders the task challenging. This paper scrutinizes the performance of naive Bayes, KNN, and logistic regression on significantly imbalanced credit card fraud data comprising 284,807 transactions from European cardholders. The dataset's skewness is addressed through a hybrid under-sampling and oversampling approach. The three techniques are applied to both unprocessed and preprocessed data. Fraud detection, defined as a set of activities thwarting illicit acquisition of assets or funds through deceptive means, varies across industries and methods. Credit card fraud, particularly susceptible due to its ease and prevalence in e-commerce and online platforms, prompted the adoption of diverse machine learning strategies to combat rising fraud rates. This paper employs machine learning algorithms for credit card fraud detection, utilizing a publicly available credit card dataset for model evaluation. While acknowledging that achieving 100% accuracy in fraud detection is elusive, the paper emphasizes the real-world applicability of its findings through the analysis of credit card data from a financial institution. In addition to assessing model efficacy, the study introduces noise into the data samples to evaluate algorithm robustness. Experimental outcomes underscore the effectiveness of the majority voting method, achieving commendable accuracy rates in detecting credit card fraud cases. The study sheds light on the pressing issue of credit card fraud, emphasizing the importance of deploying robust machine learning approaches for timely and accurate detection in real-world scenarios.