Drift
Model Drift
Progressive degradation of model performance when the real-world data distribution changes over time, differing from the original training data (Concept Drift).
Simone Disabato, Manuel Roveri
Mitigation Strategy
Continuous monitoring of performance metrics in production, automated detection of distributional shift, and establishment of periodic retraining pipelines with updated data.
Atomic Number
12
Dr
Risk ID
mg-12
Severity
5/10
Severity Level