Fraud detection model for new accounts

The challenge

Banks face attempts of fraud, including the creation of new bank accounts with the intention of defaulting payments. The Criminal Offenses team is the department responsible for detecting and blocking these fraudulent new accounts.

Objectives

They needed to learn how to detect which newly created accounts were generated with the intention of defaulting payments. The model creates a scoring system that prioritizes accounts with a higher probability of being fraudulent, enabling proactive detection and prevention of these accounts.

How have we done it?

• Machine learning
• Gradient boosting (XGBoost)
• Word2vec
• Bank transaction embedding

Results provided

The solution assists the Criminal Offenses team in prioritizing the most suspicious accounts for analysis and proactively detecting fraud. By doing so, it prevents this type of fraud from occurring, leading to a more secure and fraud-resistant banking environment.