Professional Documents
Culture Documents
I.
INTRODUCTION
PROBLEM DESCRIPTION
DATA SETS
Variables
Definition
UPDATE_COUNT
CC_PRODUCT_PRICE
OPTIONAL_FLAG
ORG_FLAG
MAIL_STATUS
CLASS
TABLE II.
Product
Low-Price
Mid-Price
High-Price
IV.
Positive
13,709
8,013
10,961
Negative
5,302
1,764
2,265
RESULTS
The SMO algorithms implement the sequential minimaloptimization algorithm for training a support vector
classifier, using kernel functions, here we used the RBF
kernel. Fig. 5 shows the output of SMO on the customer
defection data. Since the customer defection data contains
two class values, two binary SMO models have been output,
one hyperplane to separate each of the possible pair of class
values. Moreover, the hyperplanes are expressed as functions
of the attribute values in the original space [20].
Low-Price
J48
0.11
Mid-Price
0.13
5.66
4.3
299..8
High-Price
0.13
5.44
4.3
342.4
Low-Price
Mid-Price
High-Price
Accuracy
72.12%
72.28%
Recall
83.91%
84.21%
Precision
74.10%
74.14%
MLP
SMO
J48
68.81%
68.81%
81.95%
80.51%
84.93%
85.80%
72.04%
70.42%
88.14%
RF
82.32%
86.12%
88.22%
MLP
78.73%
91.32%
80.83%
SMO
82.28%
90.41%
80.92%
J48
82.87%
76.39%
92.87%
RF
MLP
SMO
83.13%
68.57%
82.71%
77.68%
67.57%
75.21%
92.61%
76.54%
91.51%
DISCUSSION
J48
Higher
Higher
Higher
High
RF
Low
Higher
Higher
High
MLP
High
Low
High
Lower
SMO
Low
Low
Low
Higher
REFERENCES
[4]
[5]
[6]
[7]
[10]
[11]
Classifiers
[3]
[9]
VII. CONCLUSION
[1]
[2]
[8]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]