gure 3.7(b) shows the QDA discrimination boundary for the data

Figure 3.6 as well. It can be seen that the QDA discrimination

was bended. The bending pushed two sides of the boundary

o the triangle class with a smaller variance. In this way, the cross

h a greater variance occupied more space than the triangle class

d a smaller variance. In other words, the bended boundary shows

ic curve. The use of this bended boundary for the data set only

wo misclassified data points. The reason that the QDA model

med the LDA model for this data set was because the QDA

mployed different covariance matrices for two classes with

variances for this data set.

Fig. 3.6. A data set with different covariance matrices in two classes.

(a) (b)

) The LDA model for data shown in Figure 3.6. (b) The QDA model for data

igure 3.6.