Data Mining: Practical Machine Learning Tools and Techniques, Second Edition

and discard the rest. In this case, three principal components account for 84%

of the variance in the dataset; seven account for more than 95%.

On numeric datasets it is common to use principal components analysis

before data mining as a form of data cleanup and attribute generation. For

example, you might want to replace the numeric attributes with the principal

component axes or with a subset of them that accounts for a given proportion—

say, 95%—of the variance. Note that the scale of the attributes affects the

308 CHAPTER 7| TRANSFORMATIONS: ENGINEERING THE INPUT AND OUTPUT

Axis

1 2 3 4 5 6 7 8 9

10

61.2%

18.0%

4.7%

4.0%

3.2%

2.9%

2.0%

1.7%

1.4%

0.9%

61.2%

79.2%

83.9%

87.9%

91.1%

94.0%

96.0%

97.7%

99.1%

100%

Variance Cumulative

(a)

percentage of variance

70%

60%

50%

40%

30%

20%

10%

0%

1 2 3 4 5 6 7 8 9 10

component number

(b)

Figure 7.5Principal components transform of a dataset: (a) variance of each compo-

nent and (b) variance plot.