Author(s): Shivansh Chandel and Bhavna Sharma

Abstract: Such issues as fraudulent operations within financial systems have been one of the most topical problems of organizations, banks, and individuals. Fraud is dynamic and its opponents keep evolving their tactics and therefore conventional systems that involve rules are becoming more and more useless. This has necessitated advanced on data-based measures on early detection and prevention of frauds. This paper is directed towards applying and evaluating various data mining and machine learning algorithms and identifying fraud in the massive transaction processing in transactional data where Kaggle Credit Card Fraud Dataset is considered an exemplary case. The analysis focuses on the working capacity of different algorithms including the Logistic regression, decision tree, random forest, support vegetable machine (SVM) and extreme gradient Boosting (XG Boost). Synthetic Minority Oversampling Technique (SMOTE) solved the high issue of imbalance which the information possessed, which serves to guarantee better sensitivity to corrupt incidences. Models were trained using stratified cross-validation, combined with hyperparameter optimization through Grid Search CV and Randomized Search CV. Experimental results reveal that XG Boost consistently outperforms other models across precision, recall, and accuracy and the F1-score, which prove to be strong enough to find a compromise between diversity in detecting the accuracy and the reduction in false positives. Random Forest also gave competitive performance as opposed to Logistic Regression and Decision Tree which displayed moderate performance. In addition to accuracy, interpretability and flexibility of models in the dynamic world of fraud are emphasized in the study. The main contributions of the work are the comparative analysis of the popular models, the effective approach to the resolution of the problem of the imbalance between classes, and understanding of striking them in real-life applications. Lastly, the constraints and research points are addressed and its focus should involve creating flexible and real-time fraud detection.

DOI: 10.33545/27076636.2025.v6.i2b.129

Pages: 207-214 | Views: 49 | Downloads: 19

Download Full Article: Click Here