Model Risk Management (MRM): Five Critical Human Interventions for ML Model Evaluations

Decision scientists should consider five critical aspects when evaluating ML models of MRM projects.

i: ML-based feature reduction approaches are not always suitable for business strategy development.

ii: Use of macroeconomic, policy, and shock variables in ML model as "Control Vars."

iii: Less important features within multilevel modeling framework may be critical to increase model performance in production.

iv: Detrending is critical when seasonal anomalies create a linear trend in time-series data.

v: Use of statistical and ML-based models simultaneously to avoid modeling failures.