Description
Logistic regression is a generalised linear model that maps features to a class probability via the logistic / softmax link. It is the calibration-friendly linear baseline in CSI sensing — interpretable coefficients, well-defined uncertainty, fast inference — and is the default head on top of CNN/Transformer feature backbones.
When it's used
- Linear baseline for hand-crafted CSI features
- Final classification layer of deep models
- Calibrated-probability estimators
Limitations
- Cannot model non-linear feature interactions without explicit basis expansion
- Underperforms ensembles on raw subcarrier features