3143. Financial Fraud Detection; Machine Learning; Data Balancing; Explainable AI; Transaction Networks
Invited abstract in session WA-38: Industrial Optimization, stream Data Science meets Optimization.
Wednesday, 8:30-10:00Room: Michael Sadler LG19
Authors (first author is the speaker)
| 1. | Victor Chang
|
| Business Analytics and Information Systems, Aston University |
Abstract
The intricacy of money laundering operations is becoming increasingly difficult for financial institutions to handle, making compliance and detection jobs extremely difficult. This study suggests a novel approach that combines resampling approaches, specific network-derived characteristics, and cutting-edge machine learning models for increased detection accuracy. Using the sophisticated, realistic synthetic transaction dataset SAML-D, we assess four cutting-edge machine learning techniques: Artificial Neural Networks (ANN), AdaBoost, XGBoost, and Gradient Boosting (GB). We thoroughly examine data balancing strategies, including SMOTE, ADASYN, Random Undersampling, and NearMiss, considering the significant class imbalance feature of financial fraud data, where fraudulent transactions make up just 0.1% of all occurrences. Furthermore, by adding network centrality measurements, the suggested method expands the feature set and successfully captures structural patterns and transaction interactions. According to experimental findings, the Gradient Boosting algorithm, in conjunction with SMOTE, performs at its best, achieving an AUROC of 0.88, a sensitivity of 0.81, and a specificity of 0.80. Despite the constantly changing landscape of money laundering tactics, financial companies can retain sufficient and comprehensible fraud detection procedures due to this integrated and flexible framework.
Keywords
- Machine Learning
- Financial Modelling
- Artificial Intelligence
Status: accepted
Back to the list of papers