Dynamic Ensemble Learning for Accurate Credit Card Fraud Detection

Abstract

 In financial organisations, credit card fraud detection is an important problem to solve because it is possible for hackers to impersonate legitimate credit card users. We tackle the class imbalance problem in fraud detection using multiple resampling strategies like oversampling, undersampling and SMOTE techniques with various datasets like European Data and Sparkov Data. Ensemble learning is used to boost the classification performance by using multiple algorithms to improve accuracy and resistance to errors. In this paper, an ensemble based framework is proposed, which employs complex resampling methods to enhance model training and prediction. A thorough analysis of several classification models reveals that a Stacking Classifier is the optimum model; it combines many models and gives better accuracy, precision, recall and F1 scores than any other strategies. Such an approach could be a substantial benefit to fraud detection systems, allowing them to detect fraudulent transactions with a high degree of confidence and at the same time reducing the number of false positives. The suggested approach emphasises the significance of ensemble methods and data balancing techniques in dealing with the complicated nature of financial fraud detection.