Fuzzy Measures and Integrals 1. 2. 3. 4. 5. Fuzzy Measure Belief and Plausibility Measure Possibility and Necessity Measure Sugeno Measure Fuzzy Integrals Fuzzy Measures • Fuzzy Set versus Fuzzy Measure Fuzzy Set Underlying Set Vague boundary Fuzzy Measure Crisp boundary Vague boundary: of fuzzy set Probability Representation Membership value of an element in X Degree of evidence or belief of an element that belongs to A in X Example Set of large number A degree of defection of a tree Degree of Evidence or Belief of an object that is tree Fuzzy Measure • Axiomatic Definition of Fuzzy Measure g : P( X ) [0,1] Axiomg1 : (BoundaryCondition)g () 0 and g ( X ) 1 Axiomg2 : (Monotonic ity)For every A, B P( X ), if A B, then g ( A) g ( B) Axiomg3 : (Continuity) For everysequence of X , if either A1 A2 A3 ... or A1 A2 A3 ... then limi g ( Ai ) g (limi Ai ) • Note: A B A and A B B, thenmax(g ( A), g ( B)) g ( A B) A B A and A B B, thenmin(g ( A), g ( B)) g ( A B) Belief and Plausibility Measure • Belief Measure g : Bel( X ) [0,1] (1) Bel is a fuzzy measure (2) Bel( A1 A2 ... An ) Bel( Ai ) Bel( Ai A j ) ... (1) n 1 Bel( A1 A2 ... An ) i i j • Note: Pr(A B) Pr(A) Pr(B) Pr(A B) and Pr(A A ) Pr(A) Pr(A ) Pr(A A ) Pr(A) Pr(A ) 1 Bel( A A ) 1 Bel( A) Bel( A ) • Interpretation: Degree of evidence or certainty factor of an element in X that belongs to the crisp set A, a particular question. Some of answers are correct, but we don’t know because of the lack of evidence. Belief and Plausibility Measure • Properties of Belief Measure 1. A B Bel( A B) Bel( A) Bel( B) 2. A B Bel( B) Bel( A) 3. Bel( A) Bel( A ) 1 Note: A It is a tree. A It is not a tree. Bel( A ) may be 0, when theinterestis given only a tree. • Vacuous Belief: (Total Ignorance, No Evidence) Bel( X ) 1 Bel( A) 0 for all A X Belief and Plausibility Measure • Plausibility Measure g : Pl( X ) [0,1] (1) Pl is a fuzzy measure (2) Pl( A1 A2 ... An ) Pl( Ai ) Pl( Ai A j ) ... (1) n 1 Pl( A1 A2 ... An ) i i j • Other Definition Pl( A) 1 Bel( A ) Pl( A ) 1 Bel( A ) 1 Bel( A) or Bel( A) 1 Pl( A ) • Properties of Plausibility Measure 1. Pl( A A ) Pl() 0 Pl( A) Pl( A ) Pl( A A ) Since Pl( A A ) 1, Pl( A) Pl( A ) 1 2. 1 Pl( A) Pl( A ) Pl( A) (1 Bel( A)) Pl( A) Bel( A) How to calculate Belief • Basic Probability Assignment (BPA) m : P( X ) [0,1] such that (1) m() 0 (2) m( A) 1 A X • Note 1. m is not equal to probability mass. 2. not necessarily m( A) m( B) even B A 3. not necessarily m( X ) 1 4. no relationship m( A) and m( A ) How to calculate Belief • Calculation of Bel and Pl Bel( A) m(B) B A Pl( A) m(B) B A • Simple Support Function is a BPA such that In X pick a subset A for which m( A) s 0 and m( X ) 1 s • Bel from such Simple Support Function s if A C and C X Bel(C ) 1 if C X 0 if A C How to calculate Belief • Bel from total ignorance m( X ) 1 and m( A) 0 for all A X Bel( X ) m( B ) m( X ) 1 Pl( A) B X Bel( A) m( B ) 0 when A X B A m( B ) m( X ) 1 A B Pl() m( B ) 0 B • Body of Evidence , m where a set of focalelementssuch thatm() is not zero. m theassigned BPA How to calculate Belief • Dempster’s rule to combine two bodies of evidence Combine Bel1 from m1 and Bel2 from m2 : m ( A )m ( B ) m( A) Ai B j A 1 i 2 j 1 K K m ( A )m ( B ) : Degree of Conflict Ai B j 1 i 2 j m() 0 • Example: Homogeneous Evidence m1 ( A) s1 m1 ( X ) 1 s1 m2 ( A) s2 m2 ( X ) 1 s2 m( A) {s1s2 s1 (1- s2 ) s2 (1 s1 )}/ 1 K ( K 0) m( X ) (1 s1 )(1 s2 ) X A X XX A A A A X A X How to calculate Belief • Example: Heterogeneous Evidence Bel1 focused on A Bel2 focused on B and assume A B X A X XX m1 ( A) s1 m1 ( X ) 1 s1 B A B B X m2 ( B) s2 m2 ( X ) 1 s2 m( A) (1- s2 )s1 A m( B) s2 (1- s1 ) m( A B) s2 s1 ( K 0) m( X ) (1 s1 )(1 s2 ) 0 if A B C s s if A B C but A C and B C 1 2 s1 if A C but B C Bel(C ) s2 if A C but B C 1 (1 s1 )(1 s2 ) if A B C 1 if C X X How to calculate Belief • Example: Heterogeneous Evidence Bel1 focused on A Bel2 focused on B and assume A B Bel( A) m( A) m( A B) s1 (1 s2 ) s1s2 s1 Bel( B) s1s2 s1 (1 s2 ) s2 (1 s1 ) s2 s1 (1 s2 ) increasing! • Example: Heterogeneous Evidence Bel1 focused on A Bel2 focused on B and assume A B m ( A)m ( B) m ( A)m ( B) s s A B 1 m( A) 2 s1 (1 s2 ) 1 s1s2 1 2 m( B ) 1 2 s2 (1 s1 ) 1 s1s2 Probability Measure • Theorem: The followings are equivalent. 1. Bel is Baysian Bel( A B) Bel( A) Bel( B) Bel( A B) 2. All of Bel' s focalelementsare singleton m( A) 0 if A 1 3. Bel Pl 4. Bel( A) Bel( A ) 1 Joint and Marginal BoE • Marginal BPA m : P( X Y ) [0,1] R is a set of focalelementsof m i.e. m( R) 0 Let RX be theprojectionof R onto X (same for RY ) : RX {x X | ( x, y ) R for some y Y } maginalB.P .A. m X ( A) m( R) for all A P( X ) R:R X A m X and mY are non - interactive iff for all A P( X ), B P(Y ) m( A B) m X ( A) mY ( B) and m( R) 0 if R A B • Example 7.2 Possibility and Necessity Measure • Consonant Bel and Pl Measure If focalelementsare nested, then theBel and Pl measuresare called consonant. m( A) m(B ) T heorem: Let (,m) be a consonantbody of evidence.T hen Bel( A B) min{Bel( A), Bel( B)} and Pl( A B) max{Pl( A), Pl( B)}. Possibility and Necessity Measure • Necessity and Possibility Measure – Consonant Body of Evidence • Belief Measure -> Necessity Measure • Plausibility Measure -> Possibility Measure – Extreme case of fuzzy measure Nec( A B) min[Nec( A), Nec( B)] Pos( A B) max[Pos( A), Pos( B)] cf ) g ( A B) min[g ( A), g ( B)] g ( A B) max[g ( A), g ( B)] – Note: 1. Nec( A) Nec( A ) 1 Pos( A) Pos( A ) 1 Nec( A) 1 Pos( A ) 2. min[Nec( A), Nec( A )] Nec( A A ) Nec() 0 max[Pos( A), Pos( A )] Pos( A A ) Pos( X ) 1 Possibility and Necessity Measure • Possibility Distribution T heorem: Everypossibility measurecan be uniquely defined by a possibility distribution, r : X [0,1], via theformula Pos( A) max{r ( x)} xA Nec( A) 1 Pos( A ) For X {x1 , x2 ,...xn }, suppose r {1 , 2 ,..., n } be thepossibility distribution such that i j if i j. Assume that Pos defined by m, theBP A. Assume A1 A2 ... An ( X ), where Ai {x1 , x2 ,..xi }. n T hatis m( A) 0 for all A Ai and m( Ai ) 1. i 1 Everypossibility measurecan be uniquely characterized by n - tuple m {1 , 2 ,.. n }, where i m( Ai ). m is called a basic distribution. Possibility and Necessity Measure • Basic Distribution and Possibility Distribution i r ( xi ) Pos({xi }) Pl({xi }) for all xi X n n k i k i i Pl({xi }) m( Ak ) k or i i i 1 , n1 0. • Ex. r (1,1,0.7,0.3,0.3,0.3,0.3,0.2) m (0,0.3,0.4,0.0,0.0,0.0,0.1) r (1,0,0,0,...,0) m (1,0,0,0,...,0) : thesmallest possibility distribution theanswer is specific xi r (1,1,1,1,...,1) m (0,0,0,0,...,1) : totalignorancem( X ) 1 and m( A) 0 for all A X Fuzzy Set and Possibility • Interpretation – Degree of Compatibility of v with the concept F – Degree of Possibility when V=v of the proposition p: V is F rF (v) F (v) • Possibility Measure PosF ( A) sup rF (v) or NecF ( A) 1 PosF ( A) vA • Example rF 0.33 / 19 0.67 / 20 1.0 / 21 0.67 / 22 0.33 / 23 A1 {21}, A2 {20,21,22}, A1 {19,20,21,22,23} Pos( A1 ) Pos( A2 ) Pos( A3 ) 1 Pos( A1 ) 0.67, Pos( A2 ) 0.33, Pos( A3 ) 0. Nec( A1 ) 0.33, Pos( A2 ) 0.67, Pos( A3 ) 1. Summary Fuzzy Measure Plausibility Measure Probability Measure Belief Measure Necessity Measure Possibility Measure Sugeno Fuzzy Measure • Sugeno’s g-lamda measure g is a fuzzy measuresatisfyingthefollowingcondition. For all A, B P( X ) with A B , g ( A B) g(A) g(B) g(A) g(B) for some 1. T heng is calledSugeno measureor g m easure. • Note: 1. 0 g 0 is a probability measure. 2. g ( A A ) g ( A) g ( A ) g ( A) g ( A ) 1 g ( A) g ( A ) 1 g ( A) g ( A ) If 0, then g ( A) g ( A ) 1 Belief Measure If 0, then g ( A) g ( A ) 1 Plausibility Measure g ( A) g ( B) - g ( A B) g ( A) g ( B) 3. g ( A B) 1 g ( A B ) Sugeno Fuzzy Measure • Fuzzy Density Function X {x1 , x2 ,..., xn } g i : g ({xi }) is called fuzzy densit y funct ion. Not e: A B , B C , A C g ( A B C ) g ( A) g ( B ) g (C ) ( g ( A) g ( B) g ( B ) g (C ) g ( A) g (C )) 2 g ( A) g ( B) g (C ) or g ({x1 , x2 , x3 }) g 1 g 2 g 3 ( g 1 g 2 g 2 g 3 g 1 g 3 ) 2 g 1 g 2 g 3 In general, X {x1 , x2 ,..., xn } n 1 n g( X ) g j j 1 n g j 1 k j 1 j g k 2 ...n 1 g 1 g 2 ...g n (1 g i ) 1 / 1 xi X Sugeno Fuzzy Measure • How to construct Sugeno measure from fuzzy density T heorem: T hefollowingequation has a unique solutionin (-1,) g ( X ) (1 g i ) 1 / 1 xi X Colloary: If {g 1 , g 2 ,...g n } is given, thenone can constructa corresponding g m easure. {g 1 , g 2 ,...g n } calculation from equation constructa g m easure. Fuzzy Integral • Sugeno Integral Definition: T heSugeno integralof a functionh : X [0,1] w.r.t.g() is g ( F ), h( x) g sup [ 0 ,1] where F x | h( x) . h(x) 1 F 2 F 1 Find t hemaximum n F n 2 . F X Fuzzy Integral • Algorithm of Sugeno Integral Reorder X {x1 , x2 ,..., xn } so that h( x1 ) h( x2 ) ... h( xn ). T hen h(x) g h( x ) g ( X ) i i X where X i {x1 , x2 ,..., xi }, and recursively given g ( X 1 ) g ({x1}) g 1 g ( X i ) g ( X i 1 {xi }) g ( X i 1 ) g i g ( X i 1 ) g i Fuzzy Integral • Choquet Integral Definition: T heChoquet integralof a function f : X [0,1] w.r.t.g() is n Cg ( f ( x1 ),...f ( xn )) ( f ( xi ) f ( xi 1 )) g ( X i ), i 1 and f ( x0 ) 0, where0 f ( x1 ) f ( x2 ) ... f ( xn ) 1 and X i {xi , xi 1 ,....,xn }. • Interpretation of Fuzzy Integrals in Multi-criteria Decision Making Set of Criteria x1 , x2 ,...xn Degree of Importance g 1 , g 2 ,...g n ObjectiveMeasurement h( x1 ), h( x2 ),...,h( xn ) Sugeno Fuzzy Integral T otalDegree of Satisfaction