Weighting
Z-BWM: Z-Number Best-Worst Method
Aboutorab, H., Saberi, M., Asadabadi, M.R., Hussain, O., Chang, E. · 2018
Overview
Z-BWM is BWM with a 'how sure are you?' channel. When you say 'Criterion B is Very Important compared to A', BWM accepts that as TFN (5/2,3,7/2). Z-BWM also asks 'how sure?': if 'High' (TFN (0.5,0.7,0.9)), it shrinks your VI-TFN by √α ≈ √0.7 ≈ 0.84. Less-sure judgments contribute less to the optimization. The crisp weight at the end uses GMIR: like a weighted-mean integrating the whole TFN shape (l + 4m + u)/6.
Strengths
- •Captures BOTH vagueness (TFN) AND reliability/confidence (second TFN) of expert judgments simultaneously
- •Demonstrably lower inconsistency (CR=0.034) vs BWM (0.382) and Fuzzy BWM (0.035) on paper's case study
- •Only 2n-3 pairwise comparisons (same as BWM): much fewer than AHP's n(n-1)/2
- •GMIR provides a single defensible defuzzification: no arbitrary choice between centroid/mean-of-maxima
Limitations
- •Fuzzy part retains subjectivity during concept translation (Aboutorab Table 4: acknowledged weakness)
- •Nonlinear constrained optimization requires solver (cannot be solved in closed form)
- •Table 3's 25 precomputed Z→TFN values are paper-specific; alternative scales require re-derivation
- •Single decision-maker: group aggregation requires external mechanism (geometric mean of Z-numbers, voting, etc.)
Method assistant
Grounded explanations: it explains the method, it does not compute.
Edge cases
- •When a_BW = (EI,VL) (i.e., u_BW=1 from Table 5), CI=3: small CI means even tiny ξ* triggers high CR; user should re-elicit
- •When all reliabilities = VH, Z-BWM reduces to Fuzzy BWM (Guo-Zhao 2017) numerically since α=1 → √α=1
- •When n=3, only one OW comparison is non-degenerate; the optimization is barely constrained: encourage n≥4
Common pitfalls
- •Don't confuse constraint scale (EI..AI, range 1..4.5) with reliability scale (VL..VH, range 0..1)
- •Don't skip Z→TFN; the optimizer needs TFNs, not (constraint, reliability) linguistic pairs
- •Don't read CI from Table 5 by the constraint level alone: index is u_BW of the Best-to-Worst TFN
- •Don't compute Σw_j = 1 on arithmetic sums of l/m/u: paper uses GMIR-normalized sum
Worked example
- 1.Step 1
- 2.Step 2
- 3.Step 3
- 4.Step 4
- 5.Step 5
- 6.Step 6
- 7.Step 7
- 8.Step 8
How to cite
Aboutorab, H.; Saberi, M.; Asadabadi, M.R.; Hussain, O.; Chang, E. (2018). ZBWM: The Z-number extension of Best Worst Method and its application for supplier development. Expert Systems with Applications. https://doi.org/10.1016/j.eswa.2018.04.015