Model Monitoring

  • Verify quality of any model at any point today and in the past
  • Verify newly implemented models as early as possible
  • Verify model stability
    • Has the score distribution changed?
    • Has the distribution of case characteristics changed?
  • Verify model ranking power
    • Do goods have high scores? Do bads have low scores? How much do goods and bads overlap?
  • Verify model calibration
    • How different are predicted values from actual values?
  • Track changes of criteria over time
Druckversion | Sitemap
© Hendrik Wagner