FRAUD DETECTION SYSTEM FOR ONLINE TRANSACTIONS USING REGRESSION ANALYSIS

The accelerated nature of online transactions has increased the risk of fraudulent activities and hence the need to have an effective and scalable detection system. In this paper, the author suggests a Fraud Detection System of Online Transactions, which is a regression-based one using Min-Max Normalization, Principal Component Analysis, Ridge Logistic Regression and is written in Python (Scikit-learn and XGBoost). The preprocessing stage provides facilitation of the descriptive homogenization of features and the PCA is efficient in diminishing the dimensionality by keeping the significant variations such that the complexity of computation is minimized. The Ridge Logistic Regression is the primary classifier used to produce understandable probability of fraud scores and regularization is used to overcome the challenge of overfitting and enhance the generalization of the model. The experimental assessment shows that the proposed system has an accuracy of 96.3, precision of 95.1, recall of 94.7, and AUC-ROC of 0.97, which is superior to the classical classifiers like decision trees, support vectors machine, and baseline logistic regression. The findings affirm the framework of real-time fraud detection and it is therefore a powerful tool to financial institutions and e-commerce sites.