Effective models depend on high-quality, informative features. Feature evaluation helps identify which features contribute meaningfully to model performance and which may be redundant or harmful.
Objectives
- Determine the relevance, predictive power, and redundancy of features.
- Guide Feature Selection, engineering, and model interpretation.
Core Aspects
Feature Relationships Assess:
- Correlation between features to detect redundancy
- Interactions that may impact the target variable
Performance Impact: Measure how feature inclusion/exclusion affects:
- Model accuracy, precision, AUC, etc.
- Stability and generalisability of results
When Evaluation is Complete
- Feature set achieves optimal model performance
- Further changes offer no significant improvement
- Feature effects are interpretable and aligned with domain knowledge