IF-MAUT - Intuitionistic Fuzzy Multi-Attribute Utility Theory (IFWA-based additive utility)

Overview

IF-MAUT extends additive utility theory to the IF setting via the Xu 2007 IFWA aggregation operator. Each alternative's row of IFN ratings is aggregated into a single utility IFN U_i, then defuzzified by the Chen-Tan score S(U_i) = μ − ν for ranking. Cost criteria are complemented at Step F2 via (μ, ν) → (ν, μ). On ties, Hong-Choi accuracy H = μ + ν breaks them.

Output

utility score, higher is better

Data

Intuitionistic Fuzzy, ifn tuples complete

Weights

Needs a weight source

Size

2+ alternatives, 3-10 criteria works best

Used for

IF-MCDM with expert hesitancy, additive utility under epistemic uncertainty, linguistic-rating decision problems mapped to IFN

How it works

1. 1

   Validate IFN axioms (μ+ν ≤ 1) on every matrix entry; verify crisp weights on the n-simplex (Σω=1). Construct the IF decision matrix R = (r_ij)_{m×n}, r_ij = (μ_ij, ν_ij).

   $$R=(r_{ij}{)}_{m\times n},r_{ij}=({\mu }_{ij},{\nu }_{ij}),{\mu }_{ij}+{\nu }_{ij}\le 1,{\pi }_{ij}=1-{\mu }_{ij}-{\nu }_{ij}$$

   Atanassov 1986 (IFS axioms); Xu 2007 §II

2. 2

   For each cost criterion j (direction='min'), complement entries via (μ_ij, ν_ij) → (ν_ij, μ_ij). Benefit criteria are unchanged. This converts every criterion to benefit-direction so that IFWA preserves preference semantics.

   $${\hat{r}}_{ij}=\{\begin{array}{ll}({\mu }_{ij},{\nu }_{ij})& \text{if }j\text{ is benefit}\\ ({\nu }_{ij},{\mu }_{ij})& \text{if }j\text{ is cost}\end{array}$$

   Atanassov 1986 IFN complement; standard IF-MADM normalisation convention (e.g. Boran 2009 Step 5 cost branch)

3. 3

   For each alternative A_i, aggregate the row of n IFNs into a single utility IFN U_i using the Xu 2007 IFWA operator with crisp criterion weights ω. The IFWA closed form (Xu 2007 Eq. 12) gives U_i = (1 - ∏(1-μ_ij)^ω_j, ∏ ν_ij^ω_j).

   $$U_i={\text{IFWA}}_{\omega }({\hat{r}}_{i1},\dots ,{\hat{r}}_{in})=\left(1-\prod _{j=1}^n(1-{\mu }_{ij}{)}^{{\omega }_j},\prod _{j=1}^n{\nu }_{ij}^{{\omega }_j}\right)$$

   Xu 2007 Def 3.3 + Theorem 3.4 (Eqs. 11-12 IFWA closed form)

4. 4

   Compute the Chen-Tan 1994 score S(U_i) = μ_{U_i} − ν_{U_i} for each alternative. S(U_i) ∈ [−1, 1] is the primary defuzzified utility used for ranking.

   $$S(U_i)={\mu }_{U_i}-{\nu }_{U_i}\in [-1,1]$$

   Chen & Tan 1994 score function; Xu 2007 Eq. (4)

5. 5

   Compute the Hong-Choi 2000 accuracy H(U_i) = μ_{U_i} + ν_{U_i} for tie-breaking on equal scores. H ∈ [0, 1]; higher H means lower hesitancy π = 1 − H.

   $$H(U_i)={\mu }_{U_i}+{\nu }_{U_i}\in [0,1]$$

   Hong & Choi 2000 accuracy function; Xu 2007 Eq. (5)

6. 6

   Rank alternatives in descending order of the (S, H) lex pair per Xu 2007 Def 3.1: primary sort by S(U_i) descending; on equality of S, break ties by H(U_i) descending; if both S and H equal, the alternatives are equivalent under IFN comparison.

   $$A_i\succ A_k⟺S(U_i)>S(U_k)\text{ or }(S(U_i)=S(U_k)\wedge H(U_i)>H(U_k))$$

   Xu 2007 Def 3.1 (score-then-accuracy comparison)

Fits when / Look elsewhere when

Assumptions to verify

• Decision matrix entries are valid IFNs (μ+ν ≤ 1)
• Preferential independence: per-criterion utility is additively separable
• Criterion weights are crisp and sum to 1 (Xu 2007 IFWA assumption)
• Expert consensus on the IFN scale (e.g. mapping linguistic terms to (μ, ν) pairs)

Edge cases and pitfalls

• •IF decision matrix R = (r_ij)_{m×n}, r_ij = (μ_ij, ν_ij).
• •tie-breaking on equal scores. H ∈ [0, 1]; higher H means lower hesitancy π = 1 − H.
• •if both S and H equal, the alternatives are equivalent under IFN comparison.

• Value-space violation: every entry must satisfy μ, ν ∈ [0,1] and μ+ν ≤ 1. Ratings elicited as linguistic terms must map to valid IFNs.

• Don't confuse IFWA (crisp weights ω) with IFWG / IFOWA / IFHA: IF-MAUT uses IFWA only. IFWG (geometric) or IFOWA (ordered) would be different algorithms, not canonical IF-MAUT.

• Operational overlap with IF-SAW: under crisp criterion weights and IFWA aggregation, IF-MAUT and IF-SAW produce identical numerical output. They differ only in framing (utility-theoretic vs aggregation-operator framing) and downstream documentation. Avoid running both on the same problem with both ranked.

Works with

How to cite