One of the best Plots for understanding relation between features
Amazing for [Features Selection], learning about useful / useless features for Models
Shows +ve & -ve correlations as numbers in a heat map
How to Choose Good Features
Collinearity: Between I/Ps and O/P:
- Don't take too much +ve correlation ( >= 0.9) -> overfitting
- -ve correlation is fine
Multi-Collinearity: Correlation between the input features too:
- Remove one of the input with too much +ve correlation.
**Example: **