Abstract
Credit card transactions Fraud is a major issue in financial sector, which is causing financial harm to both individuals
& businesses .To mitigate these risks, machine learning models have been employed to detect fraudulent transactions.
This study explores the performance of various classification models when using two different feature scaling
techniques, Standard Scaler and Min Max Scaler. The goal is to compare the effectiveness of these scaling methods in
enhancing the accuracy and precision of credit card fraud detection .In this research, we examine five commonly used
classification models: Random Forest, K Neighbors Classifier, Logistic Regression, Decision Tree Classifier, and
Gaussian NB. We assess their performance under two distinct scaling techniques: Standard Scaler and Min Max Scaler.
The analysis is carried out on a dataset of credit card transactions, where the 'Class' variable categorizes transactions
into two classes: non-fraudulent (Class 0) and fraudulent (Class 1).For each model and scaler combination, we
measure performance using key evaluation metrics, including accuracy, precision, recall, F1 score, and ROC AUC
score. Accuracy reflects the model's overall ability to correctly classify transactions, while precision indicates its
accuracy in identifying fraudulent cases. Recall measures the model's capability to capture all actual frauds, while the
F1 score provides a balanced assessment of precision and recall. The ROC AUC score evaluates the model's ability to
distinguish between classes .The findings of this research reveal significant variations in the performance of
classification models based on the choice of scaler. Notably, the Random Forest model demonstrates remarkable
performance with both Standard Scaler and Min Max Scaler, achieving near-perfect accuracy, precision, and recall
scores under Standard Scaler. This suggests that Random Forest is a robust choice for credit card fraud
detection. K Neighbors Classifier exhibits a substantial improvement in performance when using Standard Scaler,
achieving high accuracy, precision, and recall scores. Logistic Regression, too, benefits from Standard Scaler, resulting
in enhanced accuracy and precision, indicating its suitability for fraud detection tasks. Decision Tree Classifier
showcases a significant performance boost with Standard Scaler, exhibiting strong accuracy and precision, making it
another favorable choice for this application .Gaussian NB, while less accurate than other models, still demonstrates
reasonable performance, particularly with Standard Scaler. Overall, the choice of scaler plays a pivotal role in the
success of credit card fraud detection models. Standard Scaler generally leads to superior model performance,
particularly for Random Forest, K Neighbors Classifier, Logistic Regression, and Decision Tree Classifier. However,
model selection remains crucial, as certain algorithms have limitations in capturing specific patterns and classifying
cases .This study offers practical insights for financial institutions and organizations seeking to enhance their credit
card fraud detection systems. It emphasizes the importance of careful model selection and the implementation of
appropriate feature scaling techniques. The findings contribute to the ongoing efforts to combat credit card fraud
effectively, reducing financial losses and safeguarding the interests of cardholders and businesses.