IF-ENTROPY - Intuitionistic Fuzzy Entropy Weight Method (Vlachos-Sergiadis 2007 entropy measure as applied by Hung-Chen 2010)

Overview

IF-ENTROPY is an OBJECTIVE weighting method (Family = Weight_Objective): the criterion weights are derived from the information content of the decision matrix itself, not elicited from the DM. Use IF-ENTROPY when the DM cannot or does not wish to provide subjective criterion preferences. The output is a crisp weight vector on the n-simplex, which can then feed any IF ranking method (IF-TOPSIS, IF-MAUT, IF-VIKOR, etc.). The IF entropy formula (Vlachos-Sergiadis 2007) is symmetric under criterion-direction complement, so cost vs benefit criteria do not need pre-processing. Multi-DM aggregation via IFWA over DMs (Eq. 6) is built into Step F2.

Output

Weight, higher is better

Data

Intuitionistic Fuzzy, ifn tuples complete

Weights

Derived internally, no weight source needed

Size

2+ alternatives, 3-10 criteria works best

Used for

IF-MCDM weight derivation when DM cannot provide subjective preferences, Data-driven objective weighting under IFN expert ratings, Group decision making with multiple DM IFN matrices

How it works

1. 1

   Validate IFN axioms (μ+ν ≤ 1) on every matrix entry. If multi-DM input is supplied, also verify DM weights λ on the l-simplex (Σλ=1). Construct the IF decision matrix (or k matrices for group decision making).

   $$R^{(k)}=(r_{ij}^{(k)}{)}_{m\times n},r_{ij}^{(k)}=({\mu }_{ij}^{(k)},{\nu }_{ij}^{(k)}),\mu +\nu \le 1,\sum _k{\lambda }_k=1$$

   Atanassov 1986 (IFS axioms); Hung-Chen 2010 Step 1 (Eqs. 4-5)

2. 2

   (Multi-DM only; identity for single-DM) Aggregate the k individual DM matrices R^(1), ..., R^(k) into a single collective IF decision matrix D̃ = (r̃_ij) using the Xu 2007 IFWA operator with DM weights λ. For single-DM input (dm_count=1) this step is the identity.

   $${\tilde{r}}_{ij}={\text{IFWA}}_{\lambda }(r_{ij}^{(1)},\dots ,r_{ij}^{(k)})=\left(1-\prod _k(1-{\mu }_{ij}^{(k)}{)}^{{\lambda }_k},\prod _k({\nu }_{ij}^{(k)}{)}^{{\lambda }_k}\right)$$

   Hung-Chen 2010 Eq. (6); Xu 2007 IFWA operator (Def 3.3, Eq. 12)

3. 3

   Compute the IF entropy per criterion E^IFS_LT(C_j) using the Vlachos-Sergiadis 2007 measure (Hung-Chen 2010 Eq. 7). For each criterion j: sum over alternatives i of the four-term integrand [μ ln μ + ν ln ν − (1−π) ln(1−π) − π ln 2], scaled by −1/(m ln 2) so that E ∈ [0, 1]. Convention: x ln x = 0 when x = 0 (limit as x→0⁺).

   $$E_{LT}^{\text{IFS}}(C_j)=-\frac{1}{m\ln 2}\sum _{i=1}^m\left[{\mu }_{ij}\ln {\mu }_{ij}+{\nu }_{ij}\ln {\nu }_{ij}-(1-{\pi }_{ij})\ln (1-{\pi }_{ij})-{\pi }_{ij}\ln 2\right]$$

   Hung-Chen 2010 Eq. (7); Vlachos-Sergiadis 2007 (entropy measure source); Szmidt-Kacprzyk 2001 (four axioms)

4. 4

   Compute the divergence (information content) per criterion d_j = 1 − E^IFS(C_j). Higher d_j means lower entropy, hence higher information content and discrimination power.

   $$d_j=1-E_{LT}^{\text{IFS}}(C_j),j=1,\dots ,n$$

   Hung-Chen 2010 Eq. (8)

5. 5

   Normalise the divergence vector to crisp weights on the n-simplex: w_j = d_j / Σ_k d_k. The output is a crisp weight vector w ∈ Δ^{n-1} satisfying Σw_j = 1, w_j ≥ 0.

   $$w_j=\frac{d_j}{{\sum }_{k=1}^n{d}_k},\sum _j{w}_j=1,w_j\ge 0$$

   Hung-Chen 2010 Eq. (9)

Fits when / Look elsewhere when

Assumptions to verify

• Decision matrix entries are valid IFNs (μ+ν ≤ 1)
• At least one criterion has non-uniform IFN entries across alternatives (otherwise weights undefined)
• If multi-DM: DM weights λ_k crisp and sum to 1 (l-simplex)

Edge cases and pitfalls

• •If multi-DM input is supplied, also verify DM weights λ on the l-simplex (Σλ=1). Construct the IF decision matrix (or k matrices for group decision making).
• •IF decision matrix D̃ = (r̃_ij) using the Xu 2007 IFWA operator with DM weights λ. For single-DM input (dm_count=1) this step is the identity.
• •IF entropy per criterion E^IFS_LT(C_j) using the Vlachos-Sergiadis 2007 measure (Hung-Chen 2010 Eq. 7). For each criterion j: sum over alternatives i of the four-term integrand [μ ln μ + ν ln ν − (1−π

• Degenerate input: if every criterion has identical IFN entries across alternatives, then every E_j = 1, every d_j = 0, and Σd_k = 0 - division by zero in Eq. 9. Engine must guard with E-4 check. Mitigation: drop degenerate criteria or fall back to equal weights.

• Don't confuse IF-ENTROPY (Vlachos-Sergiadis 2007 measure, four-term integrand) with crisp ENTROPY (Shannon 1948 on probability normalised data, two-term -p ln p integrand). IF-ENTROPY operates directly on IFN tuples; crisp ENTROPY requires probability-normalised crisp ratios.

• Convention x ln x = 0 at x = 0 must be applied at three boundary cases in Eq. 7: μ=0, ν=0, π=0 (i.e. μ+ν=1, then 1-π=1, and ln(1-π)=ln 1 = 0 - the (1-π)ln(1-π) term vanishes naturally, no special handling needed; but μ ln μ and ν ln ν at zero require the convention).

Works with

How to cite