You are on page 1of 3

2.

Naïve Bayes

a)

Error Rate Number of instances Size

SpamBAse 20.7129 % 953+3648

SpamAssasin_modified 7.7391 % 178+2122

Precision Recall F-measure ROC

SpamBase 0.842 0.793 0.794 0.937

SpamAssasin_modifie 0.921 0.923 0.921 0.932


d

b)

no, it does not help if we only use the subset of attributes selected to construct the naïve bayes classifier
because it gives the same value as of classifying without taking subset of attributes.

c)not appropriate because each attribute describes the nature of it. So categorizing the it does not make
sense.

d)

Error Rate Number of instances Size

SpamBAse 10.15 % 467+ 4134

SpamAssasin_modified 7.7391 % 178+2122

Precision Recall F-measure ROC

SpamBase 0.899 0.899 0.898 0.964


SpamAssasin_modifie 0.921 0.923 0.921 0.932
d

1)No, it doesn’t improve the performance.

2)The number of instances count in the output is higher than the total number of observations in the
data because of the SupervisedDiscretization , it creates nominal data ( i.e. range of data) compared to
the UnsupervisedDicretization which used same output parameter as shown below.

UnsupervisedDiscretization
SupervisedDiscretization

You might also like