Fuzzy Standard Deviation (Fuzzy Standard Deviation, TFN-component profile)

Fuzzy Standard Deviation is not a ranking method; it does not order alternatives, it produces criterion weights. Instead of a single figure in each cell, it takes a triangular fuzzy number (TFN) as input: a lower bound, a most likely value and an upper bound. This represents situations in which data is not exact but given as an interval from expert judgement.

The idea is simple: the more widely alternatives scatter on a criterion, that is, the larger its standard deviation, the more weight that criterion receives. DecisionMind's profile applies this calculation separately, and independently, to each of the TFN's three components, lower bound, most likely value, upper bound, and averages the three results. Even for the crisp form of standard-deviation-based weighting, DecisionMind's own literature review has not been able to identify a single founding paper. The idea is used in common across several objective weighting methods, for example in CRITIC's spread component, but does not rest on a single source as an independent "standard deviation weighting" method. The fuzzy form of this profile is DecisionMind's own engineering decision; the manifest states explicitly that it comes from a software repository, not an academic paper.

The idea belongs to the same family as Entropy and CRITIC: if alternatives take values close to one another on a criterion, that criterion does not distinguish between the alternatives and deserves a low weight; if alternatives take values far apart, the criterion receives a high weight. Standard deviation measures this scatter directly, without an intermediary; there is no extra operation such as Entropy's logarithm or CRITIC's correlation. In the fuzzy adaptation, this measure is computed three times, separately for the TFN's three components, and the three results are averaged; the difference among the components shows how much the data's uncertainty affects the weight.

One consequence of this philosophy arises from the method's scale equalisation (min-max scaling): every criterion column is first compressed into the range 0 to 1, meaning every column's range is already equalised. The only remaining difference is how the values within a column are distributed across this range. This can make standard-deviation-based weights come out closer to one another than, say, Entropy or the Gini coefficient tends to produce; this will also be seen in the example below.

The method proceeds through three steps.

First, input validation. Every cell's triangular fuzzy number is checked to satisfy lower bound ≤ most likely value ≤ upper bound, and a direction (more is better / less is better) is confirmed for every criterion. A minimum of two alternatives and two criteria is required.

Second, three independent standard-deviation calculations. Every criterion column is scaled into the range 0 to 1 according to the criterion's direction: for a "more is better" criterion, the largest value becomes 1 and the smallest 0; for a "less is better" criterion, this is reversed. This scaling is applied separately to the TFN's lower-bound table, most-likely-value table and upper-bound table, and the standard deviation of the scaled column is computed in each table; the standard deviations are divided by their own sum, giving three separate weight vectors.

Third, averaging and an uncertainty measure. The three weight vectors are averaged and rescaled so that they sum to 1; this is the final weight vector. DecisionMind also computes and reports how far this average sits from the weight derived from the most-likely value alone, and the largest weight difference among the three components.

The formulas behind each step and the intermediate tables are given on the DecisionMind method page; this card carries no formulas.

The weight vector reflects, as the average of three fuzzy readings, how widely a criterion's scaled values scatter across the alternatives. If the difference among the components is small, the width of the TFN has changed the result almost not at all; if it is large, the choice of lower and upper bound genuinely affects the weight.

Because scaling has already equalised each column's range to 0-1, standard-deviation-based weights tend to come out closer to one another than they would under Entropy or the Gini coefficient; this is not an error but the nature of the measure, and it will be seen below in the three criteria's weights coming out close together. The weight measures a criterion's scaled scatter, not its importance to the decision-maker.

Thus instead of writing:

"The fuzzy standard deviation analysis showed that this criterion is the most important criterion"

the report should read:

"On the average of these three fuzzy readings, this criterion scatters the alternatives the most; the closeness of the weights to one another shows that the scaled columns are distributed in a similar way"

This profile works with fuzzy data (triangular fuzzy numbers, TFNs): every cell holds a triple, lower bound, most likely value and upper bound, instead of a single figure. If your data is crisp, DecisionMind's crisp counterpart of the same idea is the separate card, Standard Deviation Weighting. This fuzzy profile has no separate extension of its own within DecisionMind.

You need the following: alternatives in rows, criteria in columns, a triple satisfying l ≤ m ≤ u in every cell, and "more is better or less is better" information for every criterion. Weights are not entered; the method produces them. A minimum of two alternatives and two criteria is required. If a criterion takes the same value across every alternative (at all three components), that criterion's scaled column becomes constant and its weight approaches zero.

If your criteria are numerical but not exact, given as an interval from expert judgement, and you want weights derived from a simple, unmediated measure of scatter, without extra assumptions such as correlation or information theory, this profile is an option. Its calculation being one of the plainest fuzzy objective weighting profiles is an advantage when a quick baseline set of weights is wanted.

The cases where it should not be used follow both from its philosophy and from the nature of its source. If the relationship between criteria, whether one criterion duplicates another, needs to be taken into account, this profile cannot see it. If the result must be an academically established method resting on a single literature source, a method with a clear source should be chosen instead of this profile, since neither the crisp nor the fuzzy standard-deviation weighting has a founding paper verified in DecisionMind's literature review.

Data is fuzzy, a simple unmediated scatter measure is sufficient → Fuzzy Standard Deviation

Data is crisp → Standard Deviation Weighting

The relationship between criteria (duplication) should also be taken into account → Fuzzy CRITIC

An academic publication needs a single established method source → Choose a fuzzy weighting method with a clear source, not this profile

Weights came out too close to one another, a more distinguishing result is wanted → Try Fuzzy Gini Coefficient or Fuzzy Entropy

This profile's chief strength is its simplicity: its calculation rests only on the scatter of the scaled column, with none of PCA's correlation and eigenvalue computation or Entropy's logarithm. It is objective, quick to compute, and can be followed step by step on the table. Calculating and comparing the three components separately directly shows how much uncertainty affects the weight.

The most important limitation is the nature of its source: neither the crisp nor the fuzzy form of this profile has a single academic founding paper. DecisionMind's own manifest note states, for the crisp standard-deviation weighting, that "no founding source verified by a systematic literature review could be found"; the fuzzy form, too, is derived from DecisionMind's own source code. Second, because scaling compresses every column into the range 0 to 1, standard-deviation-based weights tend to come out close to one another; this can be uninformative for a user seeking a distinguishing set of weights. Third, running the three components independently disregards the TFN's internal consistency; three separate crisp problems are solved and averaged. Fourth, Ayan, Abacıoğlu and Basilio's (2023) review also emphasises, when comparing standard deviation with other objective methods such as Entropy and CRITIC, that it does not see the relationship between criteria or information-theoretic distinguishing power, looking only at raw scatter.

The most common mistake is presenting this profile as "the named fuzzy standard deviation method defined in the literature"; neither its crisp nor its fuzzy form has a verified founding paper.

A second mistake is interpreting the weights coming out close to one another as "the method found nothing"; this is a natural consequence of scaling having already equalised the column ranges. A third is looking only at the final weight and never examining the difference among the three components. A fourth is leaving in the table, unnoticed, a column in which a criterion takes the same value across every alternative; that criterion's weight approaches zero and it should be removed from the table.

The governing principle is this:

The fuzzy standard deviation weight is the average, across three fuzzy readings, of the scaled column's scatter; the report must show the difference among the components and must state clearly that this profile is not a literature standard but DecisionMind's own engineering profile.