Compliance & Security
Finance
Fraud Detection
Build a predictive fraud model across multiple bank locations or financial institutions without compromising consumer privacy.
Challenge
- Transaction datasets contain fewer examples of fraudulent behavior, skewing the data and undermining model performance.
- Training data is therefore needed from competitors, payment processors, and banks, but trade secrets and privacy concerns impede access.
- Model training slowed without sufficient data, giving cyber-criminals sufficient time to develop new strategies before existing models can have an impact.
Solution
- Devron allowed the credit card provider's data scientists to train fraud models on third party sets without accessing the data directly.
- Diversified data sources minimize skew and enable the company’s models to better prepare to identify edge cases and sophisticated schemes.
- Recurring model updates to incorporate data as it is generating allow fraud-detection using machine learning models to stay one step ahead in responding to cybercrime.
Outcomes
↑
Preserve Revenue
↑
Increase the Speed of Model Training
↑
↓
Minimize Risk
↓
Reduce Skew in Training Data
↓
