feature selection
- find correlations and remove uninformative features
- discretize a continuous features (binning)
- visualizations
feature transformation
- create new features
- remove dependent features
- find correlations after feature transformation
- combine features
feature scaling
- Normalization
- Standardization
- Use Standardization when -
- Unsupervised ML
- feature already has a distribution close to a Gaussian Distribution
- feature has outliers
- since normalization will squeeze all the other (mean like) values
- Use Normalization otherwise