A missing value is an absent value in data expected by an ML model. Some missingness may be acceptable, but uncontrolled levels can undermine model inputs and require a pause or fallback. section 14.3.6
Monitor missing values as an input-quality checklist:
- Measure missingness across the full dataset and separately for each feature. section 14.3.6
- Detect alternate representations such as “N/A,” “NaN,” and “undefined,” which simple checks may miss. section 14.3.6
- Inspect plots, set acceptable thresholds, and define when to pause the model or use a fallback. section 14.3.6
- Give stricter monitoring policies to important features identified through feature importance or SHAP values. section 14.3.6
