Aggregation with dependencies: Capacities and fuzzy integrals
Introduction
In this paper we outline the new trends in aggregation with inputs dependencies, and in particular we focus on capacities – normalised monotone set functions that model importances of not only individual inputs but also all their subsets, called coalitions. Capacities, also known as fuzzy measures and transferable utility (TU) cooperative games [11], [24], [38], [66], provide a powerful modelling tool to account for interactions within each coalition of inputs, and have been used for decades in Games theory and multiple criteria decision making (MCDM) [38], [40]. The aggregation is usually performed with the help of the Choquet integral [24], Sugeno integral [73] or other fuzzy integrals, see a review in [11].
The main challenge of this modelling approach is the high number of inputs coalitions (for n inputs it is ), and subsequently huge computational complexity, which involves not only the parameters but relations among them, such as monotonicity or convexity. Modern practical applications produce large universal sets, which imply the need to develop suitable representation, simplification, characterisation, eliciting, computation, and other tools. This paper reviews some of the recent developments in those areas with particular focus on large universes.
The paper is structured as follows. Section 2 outlines some of the practical aggregation problems which involve dependencies and motivate the use of capacities as a tool. Section 3 presents the capacities and fuzzy integrals. Section 4 presents the challenges when using capacities to model large scale problems and outlines some of the approaches to deal with those challenges. In Section 5 we summarise the needs for development of computational tools that would allow to successfully use those approaches. Section 6 presents the conclusions.
Section snippets
Why capacities?
The dependencies are everywhere. Aggregation in which only the individual inputs are weighted implicitly assume inputs independence, yet in reality a large number of practical problems involve inputs synergies and redundancies. Let us outline a few of such problem domains with specific emphasis on large universes.
The first area of application is the decision problems involving multiple criteria, multiple decision makers and multiple objectives. Take for example selection of a combat vehicle [62]
How they work
We consider the set of inputs , , which can be either decision criteria, optimisation objectives, portfolio items, or any other aggregation variables. Let be its power set and let denote the cardinality of a subset . We only provide some basic definitions, the rest can be found in [6], [11], [38], [40], [81].
Definition 1 A capacity (fuzzy measure) on N is a set function such that (i) , ; (ii) , implies .
The value of reflects the
Problems, challenges and approaches
While aggregation based on a given capacity seems to be straightforward, for large numbers of criteria n the exponential complexity of capacities starts to bite. There are parameters that need to be specified, and of the order of monotonicity constraints to be ensured. Not only the storage and computation with capacities becomes problematic, but to a greater degree the interpretation, acquisition, and quantification of interactions is unrealistic in practice. For example even if
Tools required
We start with the capacity representations, because depending on the representation one can achieve significant savings in terms of storage and processing capacities of a particular type. The foremost current representations are the standard and the Möbius representations, related to each other by a system of linear equations. Various interaction indices also offer alternative representations, for example the Shapley interaction index [37], [41], the nonadditivity and nonmodularity indices [81]
Conclusions
We have presented examples of applications which require modelling of interactions of large numbers of inputs. The theory of capacities, also known as fuzzy measures and cooperative games, provides suitable mathematical formalism for modelling interactions, and fuzzy integrals accommodate specified interactions at the aggregation stage. There is high computational complexity when capturing all possible interactions, which renders general capacities unsuitable for large numbers of inputs.
In this
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgement
The publication was supported by the Australian Research Council Discovery Project DP210100227 and by the RUDN University Strategic Academic Leadership Program.
References (83)
- et al.
As simple as possible but not simpler in multiple criteria decision aiding: the robust-stochastic level dependent Choquet integral approach
Eur. J. Oper. Res.
(2020) On the derivatives of set functions in matrix representation
Inf. Sci.
(2021)- et al.
Three types of monotonicity of averaging aggregation
Knowl.-Based Syst.
(2014) - et al.
Learning fuzzy measures from data: simplifications and optimisation strategies
Inf. Sci.
(2019) - et al.
Directional monotonicity of fusion functions
Eur. J. Oper. Res.
(2015) - et al.
Some characterizations of lower probabilities and other monotone capacities through the use of Möbius inversion
Math. Soc. Sci.
(1989) - et al.
Integration of genetic algorithms and neural networks for the formation of the classifier of the hierarchical Choquet integral
Inf. Sci.
(2020) - et al.
On random generation of fuzzy measures
Fuzzy Sets Syst.
(2013) - et al.
Minimals plus: an improved algorithm for the random generation of linear extensions of partially ordered sets
Inf. Sci.
(2019) - et al.
Combining analytical hierarchy process and Choquet integral within non-additive robust ordinal regression
Omega
(2016)
Methods for multi-objective optimization: an analysis
Inf. Sci.
The representation of importance and interaction of features by fuzzy measures
Pattern Recognit. Lett.
k-order additive discrete fuzzy measures and their representation
Fuzzy Sets Syst.
A review of methods for capacity identification in Choquet integral based multi-attribute utility theory: applications of the Kappalab R package
Eur. J. Oper. Res.
k-additive upper approximation of tu-games
Oper. Res. Lett.
Robust ordinal regression for value functions handling interacting criteria
Eur. J. Oper. Res.
Generalization of inclusion-exclusion integral for nondiscrete monotone measure space
Fuzzy Sets Syst.
Generalized decomposition integral
Inf. Sci.
Aggregation-based extensions of fuzzy measures
Fuzzy Sets Syst.
CC-integrals: Choquet-like copula-based aggregation functions and its application in fuzzy rule-based classification systems
Knowl.-Based Syst.
k-intolerant capacities and Choquet integrals
Eur. J. Oper. Res.
k-maxitive aggregation functions
Fuzzy Sets Syst.
Fusion functions based discrete Choquet-like integrals
Eur. J. Oper. Res.
k-maxitive fuzzy measures: a scalable approach to model interactions
Fuzzy Sets Syst.
Non-monotonic fuzzy measures and the Choquet integral
Fuzzy Sets Syst.
Capacities and overlap indexes with an application in fuzzy rule-based classification systems
Fuzzy Sets Syst.
The quadratic knapsack problem—a survey
Discrete Appl. Math.
Nonadditivity index and capacity identification method in the context of multicriteria decision making
Inf. Sci.
Nonmodularity index for capacity identifying with multiple criteria preference information
Inf. Sci.
How to build aggregation operators from data?
Int. J. Intell. Syst.
A new type of fuzzy integrals for decision making based on bivariate symmetric means
Int. J. Intell. Syst.
On random generation of supermodular capacities
IEEE Trans. Fuzzy Syst.
A Practical Guide to Averaging Functions
On representation of fuzzy measures for learning Choquet and Sugeno integrals
Knowl.-Based Syst.
Dc optimization for constructing discrete Sugeno integrals and learning nonadditive measures
Optimization
Learning Choquet integral-based metrics in semi-supervised classification
IEEE Trans. Fuzzy Syst.
Discrete Fuzzy Measures: Computational Aspects
On extended set of weights of the OWA functions
Int. J. Gen. Syst.
The axiomatic characterisations of non-modularity index
Int. J. Gen. Syst.
Learning k-maxitive fuzzy measures from data by mixed integer programming
Fuzzy Sets Syst.
Objective reduction in evolutionary multiobjective optimization: theory and applications
Evol. Comput.
Cited by (6)
Representation, optimization and generation of fuzzy measures
2024, Information FusionKnapsack problems with dependencies through non-additive measures and Choquet integral
2022, European Journal of Operational ResearchAssociated Statistical Parameters’ Aggregations in Interactive MADM
2023, MathematicsHigh-Order Non-uniform Grid Scheme for Numerical Analysis of Shortest Queue Control Problem with a Small Parameter
2023, Lecture Notes in Networks and Systems