Fig. 2From: Cost-Constrained feature selection in binary classification: adaptations for greedy forward selection and genetic algorithmsIllustration of a possible situation when applying a non-identity covariance matrix to the situation of (8). Two highly correlated normal distributed features x1 and x2, where the first component x1 has different means for the two classes (μ=1 for y=1 and μ=0 for y=0) and the second component x2 has the same mean (μ=0) for both classes. The resulting multivariate structure is perfectly separable by a linear functionBack to article page