CREDIT CARD FRAUD DETECTION USING ENSEMBLE MACHINE LEARNING METHODS

Abstract

Credit card fraud has become one of the most serious problems in modern financial systems due to the rapid growth of online transactions and digital payment platforms. Accurate and efficient detection of fraudulent transactions is important to minimize financial losses and improve the security of transactions. Traditional fraud detection techniques usually perform poorly in the presence of highly imbalanced data and evolving fraud patterns which limits their detection performance.This study provides a comparative analysis of ensemble machine learning methods for credit card fraud detection from imbalanced financial transaction data. The project implements and evaluates several machine learning algorithms like Random Forest, XGBoost, LightGBM, CatBoost and Logistic Regression to identify fraudulent transactions. In order to mitigate the class imbalance issue, the Synthetic Minority Oversampling Technique (SMOTE) is used in the preprocessing phase to improve the minority fraud cases.The proposed models are tested for various performance metrics like Accuracy, Precision, Recall, F1-Score and ROC-AUC score . The experimental results show that ensemble learning methods are significantly better than traditional classification techniques in detecting fraudulent activities. XGBoost and LightGBM show the best performance for fraud detection among the evaluated models, especially in Recall and F1-Score, which are important to reduce the number of undetected fraudulent transactions.The study results show that ensemble techniques enhance the detection of fraudulent credit card transactions significantly. The proposed framework included data preprocessing and imbalanced data handling using SMOTE and produced good results in terms of classification performance and reliable detections. Therefore, the proposed framework can be used by financial institutions as it reduces fraud and improve security. Moreover, SHAP analysis results provided transparency for the decision-making process through explaining how the predictions happened, especially which features are important in determining the fraud.