Use of Probability Density Curves

Franzen and Hillig C873 used the absolute difference of the class-conditional means m. - m. l as a criterion for feature selection. Similar methods were used by Mathews C1923 and by Pichler and Perone C2283. 

Difficulties arise if a high probability exists for zero values therefore, a feature selection method was proposed C42, 1653 that utilizes the features with most frequently encountered non-zero values (Chapter 10.2.4). 

FIGURE 46. Probability density curves of feature i for a binary classification problem. p(x. m)Ax is the probability that a pattern belonging to class m has the value of feature i in the interval Ax. 

---