- AI-powered fraud detection is enhancing the fight against VAT fraud in banking.
- VAT Carousel Fraud is a major issue in the EU, costing billions annually.
- A Dutch bank developed a machine learning model to detect this fraud.
- The challenge was creating an effective model with limited true positive cases.
- Most AML models detect a broad range of activities, but anomaly detection is effective for outliers.
- VAT Carousel Fraud has distinct patterns, especially in the missing trader role.
- A hybrid approach using supervised and unsupervised models was implemented.
- Feature engineering focused on network structures and fund movement.
- XGBoost was used to identify missing traders with supervised learning.
- Isolation Forest was used for outlier detection in unsupervised learning.
- The model is expected to perform well due to the scale and typologies of VAT fraud.
- The first models are in production, achieving a 20 percent precision rate.
- VAT carousel fraud involves exploiting the VAT system for tax evasion.
Source: zandersgroup.com
Note that this post was (partially) written with the help of AI. It is always useful to review the original source material, and where needed to obtain (local) advice from a specialist.
