Databases Reference
In-Depth Information
Actual basket:
Binary basket:
Beer
Wine
Beer
Wine
6
0
1
0
0
1
0
1
12
0
1
0
0
3
0
1
24
4
1
1
24
5
1
1
48
2
1
1
Fig. 5.
Beer, wine transactions: quantified and binary
of a co-occurring attribute. Also, some relationships may be non-linear based
on quantity [18]
Example:
Situation: Customers frequently buy either wine or beer for themselves
in varying amounts. However, when buying for a party, they often
purchase both beer and wine and they usually purchase in larger quan-
tities.
Missed rule:
When at least 24 beers purchased, wine is also purchased;
Otherwise, there is no relationship between beer and wine.
Naıvely constructing an association rule on non-quantified, binary data, in
this example, would find a rule that misleadingly represents the situation; i.e.,
Misleading rule:
When beer is purchased, wine is also purchased
{
confidence = 0.6
}
{
support = 0.43
}
This rule is misleading because it naıvely implies that purchase probabil-
ities are uniform; in fact, they are not. Under one set of conditions,
beer
and
wine
are
never
purchased together under one set of conditions; and, under
another set of conditions they are
always
purchased together.
In neither case is there a direct causal relationship. In the quantified rule
case, the larger quantities of beer and wine are caused by a third factor
(a party).
5 Describing Causality
In some ways, someone may object to this paper, as it does not offer much in
the way of solutions. It mostly identifies needs. Part of a reply is that there is
limited space and time. Another is that recognizing a need is the first step to
finding a solution. Another is that both recognizing and defining causality is
still a very complex and di
cult problem, even after over 3,000 years of effort.
Various causality descriptions and discovery tools have been suggested. It
may eventually turn out that different subject domains may have different
