Credit card fraud detection through advanced machine learning techniques

Abstract

Nowadays, digital and electronic transactions and electronic payments systems in modern days have become convenient but now it is a major challenge to face credit card fraud. Modern fraud patterns are so complex and advanced nowadays that the traditional fraud detection methods are facing difficulties to detect them. The research demonstrates how effective these patterns are for getting the high accuracy from the imbalance dataset. The implications of these results are contemporary for financial manage-ment, which offer the potential to strengthen the integrity of finances, allocate and strengthen customer trust in the face of evolving fraud threats. Multiple methods are now implemented to track the rising credit card fraud. But in this project, it determines the performance of 4 methods : Artificial Neural Networks, Support Vector Machines, Random Forest and XGBoost, using more than one dataset to evaluate their effectiveness in detecting fraudulent activ-ities. The observation includes explainable artificial intelligence (XAI) strategies. By detecting crucial elements such as transaction amount and date, the approach provides insight into versions that improve forecast transparency and reliability. The results show that combined models, particularly Random Forest and XG Boost, outperformed other approaches in terms of Accuracy, Precision, Recall, and F1-Score. The integration of LIME and SHAP adds a layer of interpretability, allowing stake-holders to understand the rationale behind the models’ decisions. This paper shows how the advanced machine learning models with explainability techniques creates a more effective and transparent fraud detection system. Including showing that XG Boost is the most effective algorithm with the highest test accuracy. Besides, this model has performed very well in accordance with precision, recall, F1-Score and accuracy than the other ML models. Despite the possibility that ANN shows strong predic-tive power in specific scenarios, its complexity limits scalability and accuracy. Here, Accuracy of fraud detection and interpretability of the models offer an efficient solution for resisting increasingly sophisticated fraud threats in the real world.