Today's article comes from the Advances in Fuzzy Systems journal. The authors are Zhang et al., from Anqing Normal University, in China. In this paper they're exploring a new strategy for solving MCDMs with fuzzy inputs. Their approach works by representing uncertain evaluations as sets of possible values, measuring the distance between those sets without distorting the data, and then deriving the importance of each criterion from the level of disagreement it creates among the alternatives.
DOI: 10.1155/adfs/6676120
Imagine that you're house-hunting. You're browsing listings, you're taking tours, you're talking to agents. All the houses you've seen so far are pretty much the same. The key difference is the neighborhood. Some neighborhoods have lower crime, some are more walkable, some are in a better school district, some are closer to transit. But on the other hand, there are others that have better restaurants and bars, better parks and green space, and more nightlife and entertainment. How do you decide? How do you choose which part of town you'd like to live in?
This is the essence of an MCDM: a multi-criteria decision-making problem. We've been studying them formally since at least the 1960s, if not earlier. The core idea is that real decisions rarely depend on a single objective. Instead, they require balancing several competing criteria simultaneously. And importantly, "solving" an MCDM is not really a thing. You do not solve it in the same sense that you solve an equation. You rank and evaluate alternatives based on the preferences you introduce into the computation. If the quality of the school district matters more to you than anything else, for example, you will get a very different "solution" than someone who simply wants to be close to the train station. It is only when the solver, or decision-maker, introduces their own priorities and weights into the model that you can arrive at a meaningful output.
Now, MCDMs are difficult enough as they are. But what happens when the criteria you want to weigh are more subjective than objective? Judging neighborhoods based on walk-score or crime-rate is relatively straightforward because those come with clean numbers that you can weigh against each other. But what about "neighborhood character"? Or "sense of community"? These kinds of variables are much harder. It is not just that they are subjective. It is that they are unclear. They are vague or imprecise, or you might just not have very much information about them at the time you're making the decision. These variables are, what we call, "fuzzy*"*.
Today's paper is a meditation on just that kind of situation. How do you find an optimal ranking for an MCDM when the inputs themselves are uncertain and even experts cannot agree on a single evaluation value? The authors propose a method called a Pythagorean Hesitant Fuzzy TOPSIS with conflict-based weighting. It works by representing uncertain evaluations as sets of possible values, measuring the distance between those sets without distorting the data, and then deriving the importance of each criterion from the level of disagreement it creates among the alternatives. If that sounded overwhelming, don't worry. On today's episode we're going to walk through what all of that means, step by step. Pythagorean hesitant fuzzy sets, the distance measure the authors propose, how it all feeds into a TOPSIS ranking method, and how conflict-based weighting actually works. Let's dive in.
First, a disclaimer: On a recent episode of Journal Club (Evaluating Sustainable Business Model Innovations Under ESG by Signed Distance Based Intuitionistic Fuzzy MCDM Method) we covered how fuzzy numbers, fuzzy sets, and intuitionistic fuzzy sets work. Today we're going to pick up where that episode left off, and go further down the rabbit hole. If you have not watched that episode, or are unsure about any of those topics, you should start there. But here is the very abridged version anyway, to make sure we are all on the same page. Three concepts you need to know.
Now, today's paper builds on top of all of that with several more concepts. The first of which is Pythagorean Hesitant Fuzzy Sets. Let's walk through those.
Up to this point, even our most sophisticated concept, intuitionistic fuzzy sets, still assumed that each evaluation ultimately resolved to one membership value and one non-membership value. But in real decision-making problems, you might not end up with a single number. One evaluator might think that the membership-value should be 0.7, another might say 0.8 or 0.5 etc. Instead of forcing those opinions into one value, a hesitant fuzzy set simply keeps all of them. So the membership value of the element in the set becomes a set of values itself. A list of possibilities rather than an average. So while intuitionistic fuzzy sets do capture uncertainty about membership, hesitant fuzzy sets go a little further: capturing the nature of the disagreement about what the membership value should be in the first place.
Then there's the Pythagorean fuzzy set. Remember: in intuitionistic fuzzy sets the membership degree and the non-membership degree, summed to no more than 1. They could sum to less than 1, sure, but they couldn't be more. Pythagorean fuzzy sets relax that constraint. Instead of requiring the sum of membership and non-membership to be less than or equal to 1, they require the sum of their squares to be less than or equal to 1. So for example:
Under this new regime, both membership and non-membership can be relatively high at the same time. That makes the representation more flexible, and better able to capture real-world situations where there may be meaningful evidence both supporting and contradicting a statement.
So what happens when you combine these ideas? When you add on "hesitant" to a Pythagorean fuzzy set?
Well, instead of assigning a single membership value and a single non-membership value, the Pythagorean Hesitant Fuzzy Set allows sets of possible membership values and sets of possible non-membership values, while still respecting the Pythagorean constraint. You still have a set of values, but every possible membership and non-membership pair in those sets has to satisfy the rule that the sum of their squares can't exceed 1. So what happens when you start to pair up different members of each set? When you pair, let's say the third value in the non-membership set to the first value in the membership set? And then the 2nd with the 3rd, and the 5th with the 2nd, etc. Well, conceptually, each evaluation becomes a small cluster of possible beliefs about how strongly an alternative satisfies a criterion and how strongly it does not. This structure makes it possible to capture both hesitation among experts and greater flexibility in how membership and non-membership interact.
Once you represent evaluations in this way, the next question becomes: how do you compare them? That is where the paper's proposed distance measure comes in. In many fuzzy MCDM methods, ranking alternatives requires computing distances between evaluations. But existing distance measures for Pythagorean hesitant fuzzy sets often require that the lists of possible values have equal length. To achieve that, researchers frequently pad the shorter lists with duplicated values. The problem is that this process artificially modifies the original information and can distort the decision outcome. Here the authors propose a new distance measure that avoids this. Instead of forcing the sets to match in size, the method compares each value in one set with the closest value in the other set and aggregates those differences. In effect, it measures how far apart two hesitant fuzzy evaluations are without altering the original data structure.
With that distance measure in place, the method now can plug into an MCDM framework. In this case they're plugging it into TOPSIS, the Technique for Order Preference by Similarity to Ideal Solution. The basic idea is simple. Imagine a hypothetical alternative that performs perfectly on every criterion. That is the positive ideal solution. Now imagine another alternative that performs as badly as possible on every criterion. That is the negative ideal solution. Every real alternative lies somewhere between those two extremes. TOPSIS ranks alternatives based on how close they are to the ideal solution and how far they are from the worst one. And here the authors compute those distances using their new Pythagorean hesitant fuzzy distance measure. The result is a ranking of alternatives that accounts for uncertainty, hesitation, and the structure of the fuzzy evaluations.
The final piece of the method is something called conflict-based weighting. In most MCDM models, each criterion must be assigned a weight that reflects its importance. Traditionally those weights are chosen by experts, but that introduces subjectivity into the process. Here, the authors derive the weights from the degree of conflict among alternatives under each criterion. The idea is that a criterion that strongly differentiates the alternatives contains more useful information for the decision. If all alternatives score similarly on a criterion, it does not help much in distinguishing between them. But if their evaluations differ, that criterion carries more decision-making power. The authors measure this conflict using the distances between alternatives' fuzzy evaluations. Criteria that produce larger overall disagreement receive higher weights. Those weights are then used inside the TOPSIS ranking procedure to determine the final ordering of alternatives.
So that's how their system works, but what did they actually do with it? How did they apply it? Well, the authors demonstrate the method using a habitat selection problem for an endangered plant species. Think back to our earlier example of trying to choose a neighborhood to live in. This is similar, but instead of a person trying to choose a neighborhood, the decision-makers here are conservation scientists trying to determine which region is the optimal place for a rare tree species to survive. The species in question is the Dabie Mountain Five-needle Pine, and the alternatives are four candidate regions in the Dabie Mountain area. Each location is evaluated across several ecological criteria, and rather than assigning precise numerical scores, biologists provide hesitant fuzzy evaluations that capture uncertainty and disagreement in their assessments. Then, they run those evaluations through their full decision pipeline: representing them as Pythagorean hesitant fuzzy sets, computing distances to ideal and worst-case environmental conditions using their proposed metric, deriving criterion weights from the level of conflict among the locations, and finally ranking the regions using the TOPSIS procedure. The end result is a prioritized list of habitats, with one region emerging as the most suitable site for conservation efforts.
So what can we learn from this study? At a narrow level, the takeaway is fairly specific: when working with Pythagorean hesitant fuzzy evaluations, the way you measure distance between those evaluations matters a lot. Padding hesitant value lists to force them into equal lengths can distort the underlying information and produce counterintuitive comparisons. By defining distance in a way that respects the original structure of the hesitant sets, the authors show that you can preserve more of the information experts actually provided. Likewise, deriving criterion weights from the level of disagreement among alternatives offers a practical way to proceed when those weights are unknown or difficult to justify.
And at a broader level, the paper highlights something important about decision science more generally. Many real-world decisions involve messy inputs: uncertain measurements, subjective judgments, and disagreement among those involved. But traditional optimization methods assume precise numbers. Approaches like fuzzy MCDM exist to bridge that gap. They do not eliminate uncertainty, but they do give us structured ways to reason about it. If you want to see the authors' full proofs for the new distance measure, the conflict analysis theorems, or the calculations for the habitat selection case, I'd highly recommend that you download the paper.