Credit Card Fraud Detection Using Machine Learning Techniques
Purpose: The rapid growth of credit fraud data and credit card fraud detection is now a challenge for machine learning algorithms. Financial fraud is increasing significantly, causing losses of billions of dollars worldwide every year. In the paper the selected techniques (artificial neural networks, decision trees and random forests) were adopted and used for credit card fraud detection. Design/Methodology/Approach: Due to the large class imbalance with fraud detection datasets, the class imbalance problem and methods for preprocessing class-imbalanced datasets are presented. ML models were applied for the SMOTE dataset and compared using the F1-Score measure. Findings: In data preparation step four approaches were considered (SMOTE, Oversampling, Undersampling, Original dataset). The F1-Score showed that SMOTE approach gives the highest value in comparison to other approaches. Practical Implications: The approach presented in the paper can be used by financial institutions to develop the system to minimize their losses and minimize the credit card risk. Originality/Value: The findings presented in the paper showed that SMOTE approach can be interesting alternative to under sampling and oversampling in data preparation step. Moreover, the comparison of the selected statistical methods showed that the random forests algorithm gives the highest accuracy.