Analyzing Factors Affecting Credit Card Fraud: Four Model-based Approach

With the proliferation of the internet, credit card fraud has become a pressing issue, leading to substantial financial losses and undermining trust among consumers. This research aims to elucidate the determinants associated with credit card fraud. By importing and cleansing two databases from Kaggle, we constructed two heatmaps for comparison, subsequently selecting the most suitable database for further analysis. We then established four models: Linear Regression model, Random Forest classifier, Logistic Regression model, and Decision Tree model. By comparing the confusion matrices and ROC curves of each model, the Linear Regression model emerged as the most proficient. Ultimately, three highly correlative factors were identified in relation to credit card fraud: High-risk country, Total number of declines per day, and 6-month chargeback frequency. The findings from this research pave the way for bolstering financial security, enhancing the efficacy of fraud detection, and thereby mitigating potential losses for consumers.