Action Rules Discovery /Lecture I/ - PowerPoint PPT Presentation

Loading...

PPT – Action Rules Discovery /Lecture I/ PowerPoint presentation | free to view - id: 680b38-OGMzY



Loading


The Adobe Flash plugin is needed to view this content

Get the plugin now

View by Category
About This Presentation
Title:

Action Rules Discovery /Lecture I/

Description:

Action Rules Discovery /Lecture I/ by Zbigniew W. Ras UNC-Charlotte, USA – PowerPoint PPT presentation

Number of Views:27
Avg rating:3.0/5.0
Date added: 30 April 2020
Slides: 23
Provided by: Zbigni
Category:

less

Write a Comment
User Comments (0)
Transcript and Presenter's Notes

Title: Action Rules Discovery /Lecture I/


1
Action Rules Discovery/Lecture I/
  • by
  • Zbigniew W. Ras
  • UNC-Charlotte, USA

2
Interestingness measure
  • Rule two conditions occur together, with
    some confidence

Data Mining Task For a given dataset D,
interestingness measure ID and threshold c,
find association E such that ID(E) gt c.

Knowledge Engineer defines c
3
Interestingness Function
Two types of Interestingness Measure
Silberschatz and Tuzhilin, 1995 subjective
and objective. Subjective measure user-driven,
domain-dependent. Include unexpectedness
Silberschatz and Tuzhilin, 1995, novelty,
actionability Piatesky-Shapiro Matheus,
1994. Objective measure data-driven and
domain-independent. They evaluate rules based on
statistics and structures of patterns, e.g.,
support, confidence, etc.
4
Objective Interestingness
Basic Measures for ? ? ? Domain
card? Support or Strength card ???
Confidence or Certainty Factor
card???/card? Coverage Factor
card???/card? Leverage card???/n
card?/ncard?/n Lift n ?
card???/card?card?
5
Subjective Interestingness
  • Rule is interesting if it is
  • unexpected, if it contradicts the user belief
    about the domain and therefore surprises the user
  • novel, if to some extent contributes to new
    knowledge
  • actionable, if the user can take an action to
    his/her advantage based on this rule

Unexpectedness Suzuki, 1997 /does not depend
on domain knowledge/ If r A?B1 has a high
confidence and r1 AC?B2 has a high
confidence, then r1 is unexpected.
Padmanabhan Tuzhilin A ? B is unexpected with
respect to the belief ? ? ? on the dataset D if
the following conditions hold B ? ? ? False
B and ? logically contradict each other A ?
? holds on a large subset of D ? A? ? B
holds which means A? ? ??
6
Actionable rules
  • Action rules suggest a way to re-classify
    objects (for instance customers) to a desired
    state.
  • Action rules can be constructed from
    classification rules.
  • To discover action rules it is required that the
    set of conditions (attributes) is partitioned
    into stable and flexible.
  • For example, date of birth is a stable attribute,
    and interest rate on any customer account is a
    flexible attribute (dependable on bank).

The notion of action rules was proposed by Ras
Wieczorkowska, PKDD00. Slowinski at al JETAI,
2004 introduced similar notion called
intervention.
7
Decision table
Action Rules
  • Any information system of the form
  • S (U, AFl ? ASt ? d), where
  • d ? AFl ? ASt is a distinguished attribute
    called decision.
  • ASt - stable attributes, AFl ? d - flexible
  • Action rule Ras Wieczorkowska
  • t(ASt) ? (b1, v1? w1) ? (b2, v2 ? w2) ? ?
    (bp, vp ? wp)(x)
  • ? (d, k1 ? k2)(x), where (?i)(1? i ? p) ?
    (bi ?AFl)

E-Action rule Ras Tsay t(ASt) ? (b1, ? w1)
? (b2, v2 ? w2) ? ? (bp, ? wp)(x) ?
(d, k1 ? k2)(x), where (?i)(1? i ? p) ? (bi
?AFl)
8
Action Rules Discovery (Tsay Ras)
Objects a b c d
x1, x2, x3, x4 0 L
x1, x3 0 L
x2, x4 2 L
x2, x4 1 L
x5, x6 3 L
x7, x8 2 1 H
x7, x8 1 2 H
Objects a b c
x1, x2, x3, x4 0
x1, x3 0
x2, x4 2
x2, x4 1
x5, x6 3
Stable Attribute a, c Flexible Attribute
b Decision Attribute d
a ?
a 0
Table Set of rules R with supporting objects
Objects b c
x1, x3 0
x2, x4 2
x2, x4 1
x5, x6 3
Objects b c
x1, x2, x3, x4
Objects a b c
x7, x8 2 1
x7, x8 1 2
c ?
Objects b
x1, x2, x3, x4
c 1
c ?
c 0
a 2
a ?
T6
Objects b c
x7, x8 1
Objects b
x2, x4 2
x5, x6 3
Objects b
x1, x3
Objects b
x2, x4
Objects b c
x7, x8 1 2
T4
T5
c ?
c 2
Figure of (d, L)-tree T2
T3
Objects b
x7, x8 1
Objects b
x7, x8 1
(T3, T1) (a 2) ? (b, 2?1) ? ( d, L ? H)
(a 2) ? (b, 3?1) ? ( d, L ? H)
T1
T2
Figure of (d, H)-tree T1
9
Application domain Customer Attrition
Facts
  • On average, most US corporations lose half of
    their customers
  • every five years (Rombel, 2001).
  • Longer a customer stays with the organization,
    the more
  • profitable he or she becomes (Pauline, 2000
    Hanseman, 2004).
  • The cost of attracting new customers is five to
    ten times
  • more than retaining existing ones.
  • About 14 to 17 of the accounts are closed for
    reasons
  • that can be controlled like price or service
    (Lunt, 1993).
  • Action
  • Reducing the outflow of the customers by 5
    can double
  • a typical companys profit (Rombel, 2001).

10
Action Rules Discovery
Decision table S (U, AFl ? ASt ?
d). Assumption a1,a2,...,ap ? ASt,
b1,b2,...,bq ? AFl, ai,1?
Dom(ai), bi,1? Dom(bi). Rule r a1,1 ?
a2,1 ? ... ? ap,1 ? b1,1 ? b2,1 ? ... ? bq,1
? d1 stable part
flexible part Question Do we have to consider
pairs of classification rules in order to
construct action rules?
11
Action Rules Discovery
Decision table S (U, AFl ? ASt ?
d). Assumption a1,a2,...,ap ? ASt,
b1,b2,...,bq ? AFl, ai,1?
Dom(ai), bi,1? Dom(bi). Rule r a1,1 ?
a2,1 ? ... ? ap,1 ? b1,1 ? b2,1 ? ... ? bq,1
? d1 stable part
flexible part Action rule rd2 ? d1
associated with r and re-classification task
(d, d2 ? d1) a1,1 ? a2,1 ? ... ? ap,1 ?
(b1, ? b1,1 )? (b2, ? b2,1)? ... ? (bq, ? bq,1)
? (d, d2 ? d1)
12
Action Rules Discovery
Action rule rd2 ? d1 a1,1 ? a2,1 ? ... ?
ap,1 ? (b1, ? b1,1 )? (b2, ? b2,1)? ... ?
(bq, ? bq,1) ? (d, d2 ? d1) Support Sup(rd2 ?
d1) x ? U (a1(x)a1,1) ? (a2(x)a2,1)?...?(a
p(x)ap,1) ? (d(x)d2). /d2-objects which
potentially can be reclassified by rd2 ? d1 to
d1/ Sup(Rd2 ? d1) ?Sup(rd2 ? d1) r ?
R, where R- classification rules extracted
from S. /d2-objects which potentially can be
reclassified by rd2 ? d1 to d1/
13
Action Rules Discovery
Action rule rd2 ? d1 a1,1 ? a2,1 ? ... ?
ap,1 ? (b1, b1,1? b1,1 )? (b2, b2,1? b2,1)?
... ? (bq, ? bq,1) ? (d, d2 ?
d1) Support Sup(rd2 ? d1) x ? U
(b1(x)b1,1) ? (b2(x)b2,1) ? (a1(x)a1,1) ?
(a2(x)a2,1) ?...? (ap(x)ap,1) ? (d(x)d2).
/d2-objects which potentially can be reclassified
by
rd2 ? d1 to d1/
14
Action Rules Discovery
Let Ud2 x ? U d(x)d2. Then Bd2 ? d1
Ud2 - Sup(Rd2 ? d1) is a set of d2-objects
in S which are d1-resistant.
Let Sup(R ? d1) ?Sup(Rd2 ? d1) d2 ?
d1. Then B? d1 U - Sup(R ? d1)
is a set of objects in S which are
d1-resistant (can not be
re-classified to class d1).
15
Action Rules Discovery
Action rules rd2 ? d1, rd2 ? d3 are
p-equivalent (?), if r/bi r'/bi always
holds when r/bi, r'/bi are both defined, for
every bi ? ASt ? AFl. Let x ? Sup(rd2 ? d1).
We say that x positively supports rd2 ? d1
if there is no action rule rd2 ? d3
extracted from S, d3 ? d1, which is
p-equivalent to rd2 ? d1 and x ? Sup( rd2 ?
d3).
16
Action Rules Discovery
Let Sup(Rd2 ? d1) x ? Sup(rd2 ? d1) x
positively supports rd2 ? d1. Confidence Conf
(rd2 ? d1) cardSup(rd2 ?
d1)/cardSup(rd2 ? d1) ? Conf(r). Conf(r
? d1) cardSup(r ? d1)/cardSup(r ?
d1) ? Conf(r).
17
Cost of Action Rule Tzacheva Ras
  • Assumption S (X, A, V) is information
    system, Y ? X.
  • Attribute b ? A is flexible in S and b1,
    b2 ? Vb.
  • By ?S(Y, b1, b2) we mean a number from (0, ?
    which describes the average predicted cost of
    approved action associated with a possible
    re-classification of qualifying objects in Y from
    class b1 to b2. Object
  • x ? Y qualifies for re-classification from b1 to
    b2, if b(x) b1.
  • ?S(Y, b1, b2) ?, if there is no action
    approved which is
  • required for a possible re-classification of
    qualifying objects
  • in Y from class b1 to b2

If Y is uniquely defined, we often write ?S(b1,
b2) instead of ?S(Y, b1, b2).
18
Cost of Action Rule
  • Action rule r
  • (b1, v1? w1) ? (b2, v2? w2) ? ?( bp, vp?
    wp)(x) ?
  • (d, k1? k2)(x)
  • The cost of r in S
  • costS(r) ??S(vi , wi) 1 ? i ? p
  • Action rule r is feasible in S, if costS(r)
    lt?S(k1, k2).
  • For any feasible action rule r, the cost of the
    conditional
  • part of r is lower than the cost of its
    decision part.

19
Cost of Action Rule
  • Assumption Cost of r is too high!
  • r (b1, v1 ? w1) ? ? (bj, vj ? wj) ? ? ( bp,
    vp ? wp)(x) ?
  • (d, k1 ? k2)(x)
  • r1 (bj1, vj1 ? wj1) ? (bj2, vj2 ? wj2) ? ?(
    bjq, vjq ? wjq)(x)
  • ? (bj, vj ? wj)(x)
  • Then, we can compose r with r1 and the same
    replace
  • term (bj, vj ? wj) by term from the left hand
    side of r1
  • (b1, v1 ? w1) ? ? (bj1, vj1 ? wj1) ? (bj2,
    vj2 ? wj2) ? ?
  • ( bjq, vjq ? wjq) ? ?( bp, vp ? wp)(x) ? (d,
    k1 ? k2)(x)

20
Class movability-index
FS - decision attribute ranking positive
integer associated with a decision value
/objects of higher decision attribute ranking
are seen as objects more preferably movable
between decision classes than objects of lower
rank/. Nj i ? N FS(dj) FS(di) ?
0. Class movability-index assigned to Nj,
ind(Nj) ?FS(dj) FS(di) i?Nj
X a b d FS
x1 a1 b1 d3 1
x2 a2 b1 d2 2
x3 a1 b2 d2 2
x4 a3 b2 d1 3
21
Class movability-index
Let Pj(i) Sup(rdj ?di) /Pj(i) all
objects in U which can be reclassified from the
decision class dj to the decision class
di Pj(N) ?Pj(i) i ? N, i?j, for any N
?1,2,,k where d1,d2,,dk are all decision
classes. Class movability-index (m-index)
assigned to dj-object x indS(x)
maxind(Nj) Nj ?1,2,,k ? x ?Pj(N)
22
Questions?
Thank You
About PowerShow.com