Feature Selection by Using Data Statistics

One may assume that features with a great variance are more important for a classification than features with a small variance. For a class-independent weighting, variance v of feature i is given by equation (95). 

Weighting fpr a supervised learning uses the ratio of an inter-class variability and an intraclass variability C148/ 1531. 

Lager values of g. indicate the relative importance of feature i. 

---