Enhancing Financial Fraud Detection Using Neural Networks, Ensemble Models, and Stacking Techniques
Author(s): Cibaca Khandelwal
Publication #: 2502035
Date of Publication: 13.06.2022
Country: USA
Pages: 1-6
Published In: Volume 8 Issue 3 June-2022
DOI: https://doi.org/10.5281/zenodo.14838647
Abstract
Financial fraud detection is a persistent challenge in the banking and e-commerce industries. This study presents a conceptual framework for enhancing fraud detection by integrating neural networks (Multi-Layer Perceptron), ensemble methods (Random Forest, XGBoost, LightGBM), and stacking techniques. By addressing challenges such as class imbalance through Synthetic Minority Oversampling Technique (SMOTE) and reducing dimensionality with Principal Component Analysis (PCA), the proposed framework improves precision, recall, and AUC-ROC metrics. The findings demonstrate that stacking ensembles outperform individual models, providing a robust solution for detecting fraudulent transactions.
Keywords: Financial fraud detection, neural networks, ensemble learning, stacking ensemble, Random Forest, XGBoost, LightGBM, SMOTE, PCA
Download/View Count: 131
Share this Article