Assignment2_answer - Model Answer for Assignment 2 1 Step 1...

Model Answer for Assignment 2
1
Step 1: Find all the confident rules for a given minimum confidence. ( Given a
minimum confidence minconf, a rule is confident if conf( x→c ) ≧ minconf, where
conf( x→c ) is the confidence of rule x→c )
Step 2: From association rules to a classifier
Let C denote the set of condent rules plus the default rule, Ø → c, where c is the
majority class in relational table T. Let size(r) denote the number of values
in rule r, and let lhs(r) denote the left-hand side of r. We assume that the values in a
rule are ordered lexicographically. Similarly, rules can be ordered lexicographically.
For example, rule x,y → 1 precedes rule y → 0 lexicographically.
Binary relations: Consider two rules r and r’ in C. We say that r is ranked higher than
r’ written as r r’, if the following conditions hold: conf(r) > conf(r’); or if conf(r) =
conf(r’), but sup(r) > sup(r’); or if sup(r) = sup(r’), but size(r) < size(r’); or if size(r) =
size(r’), but r precedes r’ in the lexicographical order of rules. We say that r is more
general than r’ (or r’ is more special than r), written r ’, if lhs(r) lhs(r’).
R
p
G
p r ⊆
CF Principle: If there are choices, the rule of the highest rank has the priority.
e can use the MCF principle to resolve the conflict in covering a case: the covering
edundant rules: Arule r in C is redundant if there is some rule r’ in C that is more
ith only non-redundant rules, we can show that left-hand sides of rules are distinct,
tep 3: Pruning the Classifier
, the MCF principle implies that the highest ranked
d
DT: Consider a non-default rule r in C. The parent of r is the rule in C that is more
M
W
rule of a given case is the rule that matches the case and has the highest possible rank.
R
general and ranked higher than r, that is, r’ p r and r’ p r
G R
W
and thus, that the generalization relationship is a lattice over left-hand sides of rules.