**8. Conclusions**

and standards. There, it is shown that one of the best‐known frameworks for addressing conditional preferences introduced by CP‐nets and TCP‐nets [3, 34] is not completely quanti‐ fied yet [35] and some other improvements need to be done in order to be used for effective

On the other hand, there is a variety of methods based on quantitative measurements with supporting of only unconditional requirements, such as the Technique for Order Preference by Similarity to an Ideal Solution (TOPSIS) method [36], Simple Additive Weighting (SAW) [37], Linear Programming Techniques for Multidimensional Analysis of Preference (LINMAP)

Many researchers have been studying the problems of business process configuration and service selection by using GA‐based solutions with different characteristics and elements.

The genetic approach for service selection and composition, proposed by Canfora et al. [22], uses one‐dimensional chromosomes and utilizes fitness function with penalty factor to select genomes and lead the convergence process to optimal solution. Similarly, GA‐based approach is proposed by Gao et al. [40], by using tree‐coded algorithm for service selection and compo‐

Furthermore, different studies and experiments are developed aimed on making comparisons of GA‐based solutions with other approaches widely used for solving optimal problems. Jaeger and Mühl [41] showed that GA‐based approach has better performance compared to Hill‐ Climbing (HC) approaches measured with both, reached overall quality and the closeness to the optimal solutions. Canfora et al. [22] conducted empirical research showing that GA‐based solutions are more scalable than Integer Linear Programming (ILP) solutions. Furthermore, they showed slower performance of GA‐based solutions which is significantly increased with

However, the use of well‐known heuristic techniques for many NP‐hard problems (including ILP) have proven limitations [27] to be applied for the problem of service selection optimiza‐ tion, since various structural and semantic constraints cannot be handled straightforward. Additionally, the problem of business process families' configuration is more complex due to

The application of genetic algorithms imposes additional concerns regarding specification of stakeholders' preferences regarding selection criteria aspects [22, 40]. Commonly used approach consider simple weighting schema (e.g., Simple Additive Weighting (SAW) [42]) for defining coefficients in the fitness function, which does not respect real‐world scenarios and the needs for complex weighting mechanism for the ranking and prioritizing stakeholders'

simultaneous selection of activities for which desired services should also be selected.

ordering of decision outcomes.

larger number of available options.

sition.

requirements.

[38], Complex Proportional Assessment (CORPAS) [39], etc.

**7.3. Configuration of business processes and service selection approaches**

212 Applications and Theory of Analytic Hierarchy Process - Decision Making for Strategic Decisions

In this chapter, we proposed a novel framework which takes into account various requirements and preferences of users to produce a final optimal selection over available options. The framework is evaluated in domain of business process models configurations and obtained relative distance to optimal solution is 10.41%. The *OptSelectionAHP* approach combines novel selection criteria model for representing different kinds of preferences and overall selection criteria measurements, with dynamic penalty factors for both structural constraints and the stakeholders' preferences, to obtain fast convergence of genetic algorithms. The definition of domain‐dependent quality characteristics [29] with presented optimized search process, enable the scalability of our solution up to potential use in different domains and scenarios.

Proposed framework have potentials to be used in different domains, ranging from software engineering and problems of optimal services selection, to learning environments [43] and problem of optimal learning paths creation [44].

In our future work, we will investigate how to develop user friendly application implementing developed *OptSelectionAHP* framework, thus enabling practical use in different domains for resolving problems of optimal selections over available options. Also, our empirical work will be aimed on analyzing sensitivity of GA operators (e.g., different repair operators [45] etc.) and comparison with other approaches as alternative to meta‐heuristic approach (e.g., integer linear programming [46], etc.).
