Human Interpretable Machine Learning for Credit Risk Scoring for one of the 10 biggest Banks Worldwide


Achievement

  • Timeline: 8 months
  • Saving: 8% of 1.18 billion Euro (cost of risk)
  • Implementation in test-environment

Summary

Implementation of state-of-the-art machine learning and machine learning interpretation for credit scoring (application scoring, behavioral scoring) – resulting in significant performance improvement (GINI increase of more than 23%) while delivering better visibility through model explanation for one of the top 10 biggest banks worldwide. Implementation involved requirement analysis, development of a machine learning and machine learning interpretation framework, hand-over and training.


 

Technology Stack

  • Google Cloud Services
  • Python, R
  • H2O-framework
  • Tensor-flow, Scikit-Learn, XgBoost
  • Data Exploration, Analysis, Visualization