ARTASI (Alternative Ranking Technique based on Adaptive Standardized Intervals)

ARTASI (Alternative Ranking Technique based on Adaptive Standardized Intervals) is a ranking method for when you hold a decision table filled with numbers and want the alternatives placed in a single order. Its output is a utility score for every alternative and the rank that score produces. ARTASI does not generate weights, it takes them from outside. Pamučar, Simić, Görçün and Küçükönder proposed the method in 2023; the paper appeared in print in 2024. Shortly afterwards, Kara, Yalçın, Kaygısız, Simić, Örnek and Pamučar (2024) extended the method into a fuzzy setting and used it together with the CIMAS weighting method to evaluate website performance. The method is still new. Its application literature is small but growing quickly.

The idea behind ARTASI is to place a criterion column not into a fixed [0,1] interval, but into an interval derived from that column's own maximum and minimum values. This interval widens according to the number of alternatives; that is where the word "adaptive" comes from. In a table with few alternatives the interval stays narrow, and in one with many alternatives it widens. Normalisation thereby becomes sensitive to the size of the table.

ARTASI's second idea, much as in TOPSIS, is to score every alternative on both its closeness to an ideal reference and its distance from an anti-ideal reference. But ARTASI combines these two components not into a single distance measure but into a sum and a balancing term. This feature makes the method compensatory: a weakness on one criterion can be papered over by strength on another. The philosophical outcome is close to that of TOPSIS, the idea of "the alternative that most resembles the ideal." But ARTASI defines this resemblance not on a fixed scale but on one adapted to the data.

The method proceeds through seven steps.

First, the adaptive interval bounds are set. For every criterion column, the method adds and subtracts a small margin, dependent on the number of alternatives, to and from the column's own maximum and minimum values, producing an upper and a lower bound. Because these bounds are derived from the column's own values, they are specific to each table.

Second, first-level standardisation is performed. The method linearly rescales every cell, between these adaptive bounds, into a predetermined fixed interval, for example between 1 and 100.

Third, direction correction is applied. For benefit criteria, the standardised value is left as it is. For cost criteria, the method inverts the value using the column's maximum and minimum. This way, a large number also means "good" for "lower is better" criteria.

Fourth and fifth, ideal and anti-ideal utility are computed. The method divides every cell by its own column's maximum and multiplies by the weight, producing "utility from closeness to the ideal." It also divides the same cell by the column's minimum, producing "utility from the anti-ideal."

Sixth, the totals are summed. The method sums every alternative's ideal-closeness utilities among themselves, and its anti-ideal utilities among themselves. This produces two totals.

Seventh, the final score is derived. The method adds the two totals together and adds to this a correction term that balances the relative share of the two totals. This correction is governed by an internal parameter that lets the decision-maker set how much weight to give to closeness to the ideal against distance from the anti-ideal. DecisionMind holds this parameter, and the fixed upper and lower bounds of the standardisation interval, at default values, and reports both.

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

The final score tells you an alternative's combined closeness to the ideal and distance from the anti-ideal, relative to the other alternatives in this analysis; it says nothing more than that. The score is not a percentage and cannot be compared across different analyses, because the adaptive intervals are built afresh from each analysis's own alternative set.

Two alternatives' scores can come out exactly equal by coincidence. This is not a computational error. It usually stems from a symmetry in the table: two alternatives' values on two criteria may be mirror images of each other. When such an equality appears, the report must state it explicitly and, if necessary, propose an additional distinguishing criterion.

Thus instead of writing:

"ARTASI found the best alternative"

the report should read:

"With these weights and this alternative set, the alternative with the highest ARTASI score is this one; if scores are equal or very close, an additional distinguishing criterion may be needed"

ARTASI works with crisp data: one number per cell. DecisionMind carries three ARTASI family members alongside the base method: classical ARTASI and two picture-fuzzy extensions. One of these extensions is paired with CIMAS weighting. Which one fits your data situation is explained on the relevant data-type cards.

You need: alternatives in rows, criteria in columns, one number per cell, and no empty cells. Direction information for every criterion and criterion weights summing to 1 are also required. ARTASI does not produce weights, it takes them from outside. A minimum of two alternatives and two criteria is required; three to twelve criteria work comfortably. The adaptive interval is built from the column's own maximum and minimum values. So when an alternative is added to the table afterwards, not only that alternative's score changes but the interval itself changes too. This resembles the shift of the ideal/anti-ideal points in TOPSIS, but in ARTASI it is also felt at the normalisation stage.

ARTASI is a suitable choice if your criteria can be measured numerically, your table is complete, and you accept compensation. It is especially useful where the number of alternatives may vary from small to large. In such situations you want normalisation to be sensitive to that variation; the adaptive interval accounts for exactly this.

There are also situations where ARTASI should not be used. Do not use it if you will accept no compromise at all on one criterion. Do not use it either if you add and remove alternatives from the table frequently and do not want the interval recomputed every time; a fixed-scale method may be more stable in that case. Nor should you use it if you cannot offer a justification for the method's internal correction parameter, because if that parameter is chosen arbitrarily, the result becomes arbitrary too.

A numerical table, compensation accepted, normalisation sensitive to the number of alternatives is wanted → ARTASI

Same goal but the data is fuzzy → ARTASI's picture-fuzzy extension

A fixed-scale, more established ranking method is preferred → TOPSIS, SAW

No compromise on one criterion, sub-threshold alternatives must be screened out → screening first, then ranking

Not a ranking but weights are needed → AHP, BWM, SWARA, CIMAS (subjective); Entropy, CRITIC (objective)

ARTASI's most important feature is that it adapts normalisation to the table: the interval bounds are derived, in every analysis, from the column's own values and the number of alternatives, which makes the method flexible across problems of different sizes. ARTASI computes closeness to the ideal and distance from the anti-ideal separately before combining them, offering a more detailed comparison than TOPSIS's single distance measure. Its computational burden is moderate, and the steps can be followed on the table.

Its limitations stem both from its structure and from being young. First, classical ARTASI requires crisp data; it cannot capture the uncertainty in expert judgement. This gap is the justification for the picture-fuzzy extension published shortly after the method was proposed (Kara, Yalçın, Kaygısız, Simić, Örnek and Pamučar, 2024). Second, the internal correction parameter in the final score determines whether the decision-maker prioritises closeness to the ideal or distance from the anti-ideal. If this parameter is changed without justification, the result changes without justification too. Third, the adaptive interval is built from the column's own data. As a result, in symmetric situations, such as two alternatives being mirror images of each other on a pair of criteria, scores can come out exactly equal. This has been observed in DecisionMind's own validation test (see Case 1). Fourth, because the method was proposed in 2023, independent rank-reversal studies have not matured as far as those for TOPSIS or VIKOR.

The most common mistake is marking criterion direction wrongly; in that case the direction correction runs in reverse and the ideal alternative is built from the worst values.

A second mistake is adjusting the internal correction parameter by trial and error to obtain a "better" result and then reporting it as though it were principled. The parameter should reflect a preference, whether priority is given to closeness to the ideal or distance from the anti-ideal, and it must not be used as a tool to engineer a desired ranking. A third mistake is ignoring an exact tie between two alternatives' scores and arbitrarily declaring the first one "the winner." A tie must be stated in the report and an additional criterion proposed. A fourth mistake is comparing old and new scores after an alternative has been added to the table without accounting for the fact that the adaptive interval has also changed. A fifth is reading the score as though it were a percentage.

The governing principle is this:

An ARTASI result is a balance of ideal-closeness and anti-ideal-distance adapted to that table's own scale. Adding an alternative to the table, or changing the internal correction parameter, can change this result; the report must show this.