Smart Solutions for Utilizing Advanced Machine Learning for Robust Credit Card Fraud Detection

This research primarily aims to shed light on the serious issue of credit card fraud, which has become much worse with the advent of internet shopping and, more specifically, the present COVID-19 epidemic. Developing a machine learning system capable of distinguishing between legitimate and fraudulent credit card transactions is the primary objective of this project, which aims to decrease an annual loss of ∃24 billion. Using data such as transformed numerical characteristics after PCA analyses, transaction time and amount, and Logistic Regression, Decision Tree Classifier, and K-Nearest Neighbours approaches are employed in the research study. The accuracy rates of these algorithms are demonstrated by the cross-validation score, ROC AUC score, and F1 score within the context of fraud detection. Improving the models’ accuracy and resilience is as simple as using statistical tests like ANOVA when selecting features. To improve the detection of fraudulent behaviour and to accurately compare the results of various algorithms, balanced datasets are essential, as this shows. There was a 91% F1 score, 92.35% ROC AUC, and 98.01% crossvalidation accuracy rate for fraud detection in the logistic regression model. Alternatively, a decision tree classifier's fraud detection cross-validation score was 96.67%, ROC AUC was 91.36%, and F1 was 90%. When it came to detecting fraud, K-Nearest Neighbours performed exceptionally well with scores of 97.63% for ROC AUC, 99.34% for cross-validation, and 97% for F1.