A Pragmatic Approach to Summarize Association Rules in Business Analytics Projects - Technologies and Applications of Artificial Intelligence - page 327

summary with consequent 'F', and so on. Such ranking information allows users to
identify more important and interesting rules.
Similarly, the antecedents in individual rule summaries are ranked based on
antecedent frequency. In the rule summary with consequent 'I' when 10A60C was
set, antecedents 'E', 'M', and 'A' were the most important, this is followed by 'L' and
'Q', and lastly 'F' is the least important. Such ranking suggests affinity of the
antecedents to the consequent within a rule summary.

Table 3. Rule Summaries for the Online Purchase dataset. Confidence threshold was fixed at
60%. Antecedent Support threshold was gradually reduced from 20% to 5%.

Threshold
Setting
( x A y C)

Rule
Reduction to
Summary

Support
Range (%)

Confidence
Range (%)

Consequent Antecedent

5 rules
reduced to 2
summaries

I*3 E*1, M*1, L*1 24.36 - 34.21 62.76 - 68.77

20A60C

A*2 G*1, J*1 29.64 - 30.79 62.76 - 66.59

6 rules
reduced to 2
summaries

I*4 E*2, M*1, A*1, L*1 19.36 - 34.21 62.76 - 68.77

15A60C

A*2 G*1, J*1 29.64 - 30.79 61.93 - 66.59

E*4, M*4, A*4, L*3,
Q*3, F*1

I*11 10.64 - 34.21 60.71 - 69.61

19 rules
reduced to 3
summaries

10A60C

A*6 G*3, J*3, Q*2, O*1 11.43 - 30.79 60.12 - 69.61

F*2 H*1, N*1, I*1 12.29 - 12.71 65.17 - 70.35

E*11, M*10, A*10,
Q*8, L*6, F*3, B*2,
C*2

I*24 5.00 - 34.21 60.71 - 89.19

G*8, J*6, Q*3, O*3,
E*2, K*2, L*2, P*2,
I*1, M*1, N*1

A*17 5.57 - 30.79 60.12 - 77.89

N*4, H*3, I*3, Q*2,
A*1, B*1, L*1

F*7 5.71 - 12.71 60.00 - 84.09

A*2, K*2, J*1, M*1,
O*1, P*1

G*5 5.64 - 9.86 68.52 - 100.00 61 rules

reduced to 10
summaries

5A60C

J*2 A*1, G*1, M*1, O*1 6.93 - 7.71 60.19 - 62.89

K*2 P*2, G*1 6.14 65.12

E*1 A*1, I*1, L*1 7.86 60.00

L*1 F*1, H*1 8.64 61.16

N*1 F*1, J*1 5.86 84.15

Q*1 B*1, N*1 7.57 61.32

Note that the ranking of rule summaries tends to remain the same even though
there are changes in the antecedent threshold settings. Similarly, the antecedent
ranking within each summary is also more-or-less retained despite changes of the
antecedent support threshold settings. These observations suggest that the rule
summaries tend to be consistent despite the changes in the antecedent support
threshold settings.