Database Reference
In-Depth Information
confidence of an association rule is defined as support ( X Y )
support ( X ) , and measures the
confidence with which an antecedent can accurately predict the consequent. The
lift of an association rule combines the previous two measures and is defined as
conf idence
support ( Y ) . As such, the lift takes both the confidence and the representativeness
(support) of an association rule into account.
We used the popular WEKA data-mining platform (version 3.6.5) for a very
succinct mining exercise. A preliminary run of the Apriori algorithm for the
10 rules with the highest confidence (minimum support of 0.01) shows the
following output:
1. clothes F 3=true clothes F 4=true 238
clothes MF=true 201
conf:(0.84)
2. clothes F 4=true clothes knitting=true 267 clothes MF=true 217
conf:(0.81)
3. clothes F 4=true clothes F 5=true 220
clothes MF=true 174
conf:(0.79)
4. household 3=true clothes F 4=true 258
clothes MF=true 199
conf:(0.77)
5. shoes=true clothes knitting=true 221
clothes MF=true 169
conf:(0.76)
6. clothes F 1=true 241 clothes MF=true 180 conf:(0.75)
7. clothes M 1=true 385 clothes MF=true 281 conf:(0.73)
8. clothes F 4=true 1089 clothes MF=true 777 conf:(0.71)
9. household 1=true shoes=true 236 clothes MF=true 168 conf:(0.71)
10. clothes F 5=true 414 clothes MF=true 293 conf:(0.71)
The first important point to notice is that all rules contain clothes FM as an
item in their consequent. In fact, 54 out of the 64 rules found in total (minimum
support of 0.01, minimum confidence of 0.3) contain this item. As this anchor
store in the shopping mall attracts the majority of visitors (see Table 14.1 ), it
appears in a large number of rules with high levels of confidence and hence also
pollutes the view with rather obvious rules. Accordingly, we removed this store
from the data set and reran the algorithm (minimum support of 0.005, sort by
lift with a minimum lift of 1.1) in order to mine for less obvious (and hence
more interesting) rules. The algorithm finds 266 rules, out of which the 20 top
rules are shown below:
1. clothes F 3=true clothes F 4=true 238
clothes F 5=true 88
conf:(0.37) < lift:(14.9) > lev:(0) [82] conv:(1.54)
2. clothes F 5=true 414
clothes F 3=true clothes F 4=true 88
conf:(0.21) < lift:(14.9) > lev:(0) [82] conv:(1.25)
3. clothes F 4=true 1089
clothes F 3=true clothes F 5=true 88
conf:(0.08) < lift:(10.62) > lev:(0) [79] conv:(1.08)
 
Search WWH ::




Custom Search