Description
Decision Trees recursively partition feature space along axis-aligned splits chosen by an impurity criterion (Gini, entropy, MSE). They are interpretable, fast at inference, and the building block of random-forest and gradient-boosting ensembles. In CSI sensing they appear mostly as leaf-level baselines or as model-explanation aids when feature attribution matters.
When it's used
- Interpretable single-model baseline
- Threshold-rule extraction from CSI feature pipelines
- Building blocks of ensemble methods (RF, gradient boosting)
Limitations
- High variance — small data shifts produce different trees
- Cannot capture diagonal decision boundaries efficiently
- Underperforms ensembles on almost every CSI benchmark