An Evaluating Method with Combined Assigning-Weight Based on Maximizing Variance
This paper proposes a combined assigning-weight approach to determine attribute weights in the multiattribute decision problems. The approach combines subjective weights and objective weights of attributes based on maximizing variance. Objective weights are determined by rough set method and subjective weights by Analytic Hierarchy Process. This new combination method may integrate the merits of both subjective and objective weighting methods. Empirical study shows that the new method can lead to more reasonable weighting results and decision.
During the process of multiattribute decisions, it is necessary to establish a set of index systems for evaluating attributes and determine relative weights for the attributes. The weights of attributes represent the relative importance among them . They also reflect status and effect of each attribute during the process of evaluation and decision-making. Determining weights of indexes is related to reliability and correctness of order on schemes. Therefore, it has been a hotspot in the field of decision-making and evaluation theory and multiple attribute decision-making has received a great deal of attention from researchers in many disciplines . Generally speaking, there are two kinds of methods determining attribute weights including subjective assigning-weight and objective assigning-weight . In the strict sense, there is no absolute objective assigning-weight method. The standard is whether the weights are determined by experts or not. In fact, even if it is an objective method, there are also some subjective factors in the entire process of a method.
Objective assigning-weight method takes advantage of information of each attribute to determine weights of attributes. It does not depend on subjective judgment of decision-makers to assign the weights. The weights are assigned to attributes by a mathematical method directly. Objective methods contain entropy evaluation [4, 5], principal component analysis , similarity scale based on alternative , goal programming [8, 9], centroid method , satisfactory degree of alternative , two-phase method , case-based reasoning , genetic algorithms , rough set [15, 16], fuzzy preference , interval-valued intuitionistic fuzzy decision , and so forth.
Subjective assigning-weight method gives weight to attributes according to experience of decision-makers and subjective importance to each attribute . Subjective weights reflect accumulated experience of decision-makers and subjective judgment to present decision-making background. Variability of environment requires us to grasp importance of each attribute neatly when we make a decision. Thus, subjective methods can solve the problems. They include expert survey , dare score , importance ranking of attributes , point estimation , binomial coefficient , and judge matrix .
Determining weights by subjective weight method can utilize decision-makers’ experience and knowledge. But its flexibility and mutability lead to much more subjective randomness. Hence, it is very important to avoid subjective randomness while giving play to its advantages. At the same time, objective weight method is inconsistent with actual importance of attributes sometimes because it does not consider decision-makers’ subjective willing. Besides, it is difficult to give clear explanation to the results. Considering advantages and disadvantages of subjective and objective method, respectively, many scholars brought forward methods combining subjective and objective weights , interactive assignment , combined TOPSIS [27, 28], and so forth. However, how to choose methods determining subjective and objective weights or what methods are appropriate to combine the two kinds of methods for overcoming the shortages which occur in either a subjective approach or an objective approach is still argued. In this paper, we will study a combined assigning-weight method based on maximizing variance to determine attribute weights, where objective weights of attributes are determined by rough set theory and subjective weights by AHP. Lastly, empirical study will demonstrate the feasibility of our method.
The remainder of the paper is organized as follows. Determining subjective weights for attributes based on Analytic Hierarchy Process is introduced in Section 2. Determining objective weights for attributes based on rough set is introduced in Section 3. Combined assigning-weights for attributes based on variance maximization is described in Section 4. Empirical study is presented in Section 5. Finally, conclusive results are drawn in Section 6.
2. Determine Subjective Weights for Attributes Based on Analytic Hierarchy Process
Analytic Hierarchy Process (AHP) was put forward by American Operation Researcher Sasty in the mid 1970s, which is a simple and practical method of system analysis and evaluation combining qualitative and quantitative analyses. The basic thinking to process decision-making problems is to simplify decision-making problems with multiobjective, multicriteria, and difficulty to be standardized into single target problems with multihierarchies, calculate the importance of each element in the same hierarchy to that in upper hierarchy by pairwise comparison, and acquire the weights of indexes finally. Basic steps of evaluating problems for AHP are shown as follows.
Step one: establish index hierarchy-structure. Divide indexes on a problem into different hierarchies, and determine the relation among indexes of every hierarchy.
Step two: construct judge matrix . In multiple attributes, if there are many evaluating indexes, it is very difficult to determine the importance of each index directly. AHP requires decision-makers to judge relative importance of every index in a hierarchy and calibrate number for the judgment to form judge matrix.
Step three: calculate sequence weights. Calculate the importance of pairwise indexes for judge matrix and obtain weight of each index in the same hierarchy.
Step four: consistency check. When comparing the importance of pairwise indexes, it is not possible to meet consistency requirements because of some errors. In order to make sure of the feasibility of weight determination in AHP, consistency check needs to be done for judging results.
There are two core ideas in AHP. One is that indexes are divided into different hierarchies, namely, by establishing a hierarchy-structure model of a problem, transferring a complex problem use to a sequence calculation in hierarchies. Two is that AHP solves the problem that it is difficult to judge relative importance of multiple factors according to the idea of pairwise comparison. Because AHP can solve the sequencing problem of indexes in complex system, it is widely used in many fields.
3. Determine Objective Weights for Attributes Based on Rough Set
Rough set was a theory of data analysis brought forward by Polish Mathematician Professor Pawlak in 1982. By utilizing rough set method to analyze decision-making table, we can evaluate the importance of given attributes, construct the reduction of attribute set, and eliminate redundant attributes from decision-making table. At the same time, we also can produce clustering rules from decision-making table and apply it to make decisions.
Since Professor Pawlak proposed the theory of rough set, it has been widely used to process uncertain, imprecise, and incomplete information in the fields of artificial intelligence, cognitive science, and especially intelligent information processing as an effective mathematical tool. In recent years, some scholars have been introducing rough set theory into decision-making science for determining weights of all kinds of index objectively according to the characteristics that rough set can acquire property involved in data themselves while it does not depend on a priori knowledge . When assigning weights in decision-making of management, rough set theory is widely used and developed in many fields.
3.1. Importance of Attributes
In rough set theory, is defined as an information system, where is a discourse domain and is a condition attribute set. is an information function which endows the attribute of every object with an information value, where , , and denotes the range of attribute . Subset of every attribute decides a binary indistinguishable relation :
The relation composes one partition of , denoted by , simply denoted by . Any element is named as an equivalence class in .
A decision-making table is defined as , where is a discourse domain, is a condition attribute set, is a decision-making attribute set, is an information function, , , , and denotes a range of attribute . If decision-making attribute is removed in decision-making table, the table will become an information system, which means the difference between information system and decision-making table is whether decision-making attributes are involved or not.
In decision-making table , for , suppose that expresses the partition of condition attribute set to discourse domain . , noted as , is called as lower approximate set on of in .
In decision-making table , for , suppose that expresses the partition of condition attribute set to discourse domain and denotes the partition of decision-making attribute set to discourse domain and callas positive region on decision-making attribute set of condition attribute set in discourse domain .
In accordance with the above-mentioned definitions, we can define the concept for importance of attribute . In information system , for , assuming , importance of attribute is defined as follows:
3.2. Determine Attribute Weights
In discernibility matrix based , denotes the total amount of discernibility elements produced by attribute in discernibility matrix. Total amount size of discernibility elements produced by attribute is conducted as heuristic information; namely, if the bigger is the total amount size of discernibility elements produced by an attribute , it means the higher is important level of attribute . In our paper, it is also used to measure importance of attributes in information system.
In information system , for , define weights of attribute in information system as the following equation :
Evidently, the attribute weights are only determined by (3) rather than experts. Therefore, it is an objective method.
4. Method Combining Subjective and Objective Assigning-Weight Based on Maximizing Variance
Assuming a decision-making problem with multiple attributes, its scheme set is , and is an attribute set. , ; , is the value of scheme to attribute . is the result of standardizing decision-making matrix . If weight vectors are determined, we can calculate evaluating result of every scheme as follows :
Subjective and objective assigning-weight methods have their advantages and disadvantages. In this paper, we first use rough set theory and AHP to determine objective weights and subjective weights of attributes, respectively, and then assign weights combining subjective weights and objective weights based on maximizing variance.
Suppose the subjective weights are , in accordance with rough set theory and the objective weights are , in accordance with rough set theory. In order to absorb advantages of subjective and objective assigning-weight methods, combine two kinds of weights to integrated weight , where , and are linear combining coefficients, , , and simultaneously and meet unit restriction . If subjective weights and objective weights are made certain, integrative weights depend on and . We deduce how to determine and according to maximizing variance as follows.
In multiple decision-making, if there is no evident difference for the attribute to all decision-making schemes, the attribute will have no effect on sequencing results of decision-making schemes. So, the weight of the attribute is zero. However, if there is evident difference for an attribute to all decision-making schemes, the attribute will have big effect on sequencing results of decision-making schemes. So, the attribute should be given bigger weight. In statistics, variance is an important factor reflecting difference degree. Based on the principle of maximizing variance, weight vectors should have the total variance of all attributes to all decision-making schemes to be maximized. Thus, construct the following linear programming model :
In the above model,denotes arithmetic mean of attribute values of attribute ; namely,
In addition, according to , we can calculate values of and as follows:
After getting the values of and , it is easy to obtain integrative weights . Then, normalize and acquire the result as the final value of each weight. We can calculate comprehensive evaluating result of every scheme,
Because the method combines subjective assigning-weight and objective assigning-weight, its weights both mix decision-maker’s preference and ensure objectivity of decision-making. Furthermore, combined assigning-weight method based on maximizing variance can make evaluating values of schemes more discrete, and it is available for decision-maker to make a decision clearly.
5. Empirical Analysis
There are three kinds of methods to choose location of a logistic park, including linear programming, multiple attributes evaluation, and heuristic algorithm. When making the planning of regional logistic system, there are often some alternative schemes. Decision-making of a logistic park implies choosing an optimal one from alternative schemes. Accordingly, multiple attribute decision-making is often used to evaluate schemes. Process of location decision-making for a logistic park includes selecting evaluating indexes, determining index weights, index normalization, calculating comprehensive evaluating results, and making decision. In this paper, taking logistic center selection of a city as an example, we illustrate how to apply combined assigning-weight method to choose location of a logistic park.
5.1. Construct Index Systems of Evaluation and Mark Them
Firstly, according to the factors influencing the location of a logistic park, choose four indexes including traffic environment, benefit, matching function, and development potential as criteria layer. Each of the criteria involves its decision-making indexes. There are four hierarchies for AHP. The highest layer is goal and then criteria and subcriteria; the lowest layer is alternatives in AHP hierarchies. Assume there are eight alternatives in decision-making (see Table 1).
According to these indexes, experts mark 5, 4, 3, 2, and 1 to express excellent, good, average, poor, and poorer for each index. After calculating arithmetic mean of marks from experts, we can obtain average value of each index for every alternative (see Table 2).
5.2. Determine Subjective Weights of Indexes
Subjective weights are determined by AHP. In AHP, its fundamental problem is to find largest eigenvalue and corresponding eigenvectors of judge matrix. In practice, there are two calculation methods including accurate calculation and approximate calculation. In this paper, we use sum-product method of approximate calculation. We give an example of how to calculate subjective weights of criteria hierarchy with four indexes.
Firstly, determine judge matrix and normalize it to; namely,
Then, calculate sequence weights:
Thirdly, calculate largest eigenvalue of judge matrix BW:
Fourthly, calculate consistency check CI:
Look RI up from random consistency index table, ; then calculate consistency ratio CR:
Consequently, we can judge that there is satisfied consistency with the judge matrix and calculate relative weights of four indexes in criteria hierarchy. In the same way, determine weight of each index in subcriteria hierarchy (see Table 3).
5.3. Determine Objective Weights of Indexes
Objective weight of each index is determined by rough set theory. Firstly, discretize initial evaluating value of each index. The rule is that the excellent is in [4.5, 5.0], marked by 5; good in [4.0, 4.5), marked by 4; average in [3.5, 4.0), marked by 3; poor in [3.0, 3.5), marked by 2; poorer less than 3.0, marked by 1. According to the rule, the values in Table 2 are discretized as Table 4.
In Table 2, twelve decision-making attributes are C11, C12, C13, C21, C22, C23, C31, C32, C33, C41, C42, and C43. Our aim is to calculate evaluating values of eight alternatives of location for a logistic park.
Then, calculate the importance of attribute C11 as follows:
The results of other attribute importance are shown in Table 5.
5.4. Combine Subjective and Objective Weights
According to combined assigning-weight method based on maximizing variance, assume objective weight vector and the subjective weight vector ,. Linear combination of two kinds of weights is expressed as integrate weight , where and and are linear combining coefficients. According to the principle of maximizing variance, values of and are calculated as follows:
Calculate integrate weight and normalize it. The final results of normalization for each attribute are shown in Table 7.
5.5. Evaluating Results of Location Selection for a Logistic Park
According to initial evaluating value of each index for alternatives and results of Table 7, calculate comprehensive evaluating results of each alternative by (10). The values are seen in Table 8. Because the score of alternative L3 is the highest, it can be chosen as the optimal scheme.
Most of multiattribute decision-making uses subjective or objective methods to determine weights. However, few studies brought forward methods combining subjective and objective assigning-weight. In this paper, we brought forward a combined assigning-weight method when rough set and AHP are used to determine objective weights and subjective weights of multiattributes, respectively. For objective assigning-weight method by rough set, the bigger is the total amount size of discernibility elements produced by an attribute and the higher is the important level of the attribute. But it ignores the importance of each attribute itself because rough set determines weights according to actual values of indexes. AHP is a kind of typical subjective assigning-weight method. Wherefore, the assigning-weight method combing rough set and AHP can synthesize both of their merits. In addition, compared with other combining weight methods, the method with subjective and objective weight combination based on maximizing variance is more feasible because it considers different degree of each attribute itself. Finally, empirical study demonstrates feasibility of combining weight method. When attribute weights are entirely unknown and attribute values are continuous, the combining method is an effective evaluating tool of decision-making.
Conflict of Interests
The authors declare that there is no conflict of interests regarding the publication of this paper.
The authors are grateful for the work environment and conditions provided by Changshu Institute of Technology. Liu Hongjiu would like to thank his wife and coauthor Dr. Hu Yanrong who gave him many good advices and ideas. The work was supported by the Humanity and Social Science Youth Foundation of Ministry of Education of China (14YJA630030), the National Bureau of Statistics Statistical Science Research Project (2014LY032), and the Jiangsu Philosophical and Social Science Program for Colleges and Universities (2013SJB6300001).
I. A. Curtis, “Valuing ecosystem goods and services: a new approach using a surrogate market and the combination of a multiple criteria analysis and a Delphi panel to assign weights to the attributes,” Ecological Economics, vol. 50, no. 3-4, pp. 163–194, 2004.View at: Publisher Site | Google Scholar
X. Q. Zhang, W. H. Feng, and N. N. Li, “Attribute recognition model based on entropy weight and its application to evaluation of groundwater quality,” in Applied Mechanics and Mechanical Engineering, Parts 1–3, H. H. Tan, Ed., vol. 29–32, pp. 2698–2702, 2010.View at: Google Scholar
Z. L. Tianlei, H. Zhengyou, and Q. Qingquan, “Distribution network service restoration multiple attribute group decision-making using entropy weight and group eigenvalue,” in Proceedings of the 3rd International Conference on Intelligent System Design and Engineering Applications (ISDEA '13), pp. 602–606, IEEE, Hong Kong, January 2013.View at: Publisher Site | Google Scholar
J. M. Cisneros, J. B. Grau, J. M. Antón, J. D. de Prada, A. Cantero, and A. J. Degioanni, “Assessing multi-criteria approaches with environmental, economic and social attributes, weights and procedures: a case study in the Pampas, Argentina,” Agricultural Water Management, vol. 98, no. 10, pp. 1545–1556, 2011.View at: Publisher Site | Google Scholar
Y. Q. Xia, R. Zhu, and X. Wu, “Research of approach to linguistic fuzzy multiple attribute decision-making with weights obtained based on satisfaction,” in Proceedings of the 4th IEEE Conference on Industrial Electronics and Applications (ICIEA '09), pp. 2805–2809, May 2009.View at: Publisher Site | Google Scholar
J. Wu, X. Wu, and Z. Gao, “Multi-attribute weight allocation based on fuzzy clustering analysis and rough sets,” in Advances in Computation and Intelligence: Proceedings of the 4th International Symposium (ISICA '09), Huangshi, China, Ocotober 23–25, 2009, Z. Cai and E. N. Sanchez, Eds., vol. 5821 of Lecture Notes in Computer Science, pp. 358–365, Springer, Berlin, Germany, 2009.View at: Publisher Site | Google Scholar
L. F. Wang, “A network intrusion detection algorithm based on rough set attribute-weighted clustering,” in ISTM/2009: 8th International Symposium on Test and Measurement, Vols 1-6, 2009.View at: Google Scholar
W. Wang and Z. Wang, “An approach to multi-attribute interval-valued intuitionistic fuzzy decision making with incomplete weight information,” in Proceedings of the 5th International Conference on Fuzzy Systems and Knowledge Discovery (FSKD '08), pp. 346–350, October 2008.View at: Publisher Site | Google Scholar
Q. Zhang, X. Li, and D. M. Li, “Weight determination based on priority preference in multiple attribute decision making with interval numbers,” in Proceedings of the Chinese Control and Decision Conference (CCDC '11), vol. 1–6, pp. 2232–2235, IEEE, Mianyang, China, May 2011.View at: Publisher Site | Google Scholar
V. Marshall, E. Tong, E. Behr, and S. A. Shakir, “The epidemiological case-control results from the Drug-Induced Arrhythmia Risk Evaluation (DARE) study,” Pharmacoepidemiology and Drug Safety, vol. 21, pp. 266–267, 2012.View at: Google Scholar
D. Weissman-Miller, “Novel point estimation from a semiparametric ratio estimator (SPRE): long-term health outcomes from short-term linear data, with application to weight loss in obesity,” International Journal of Biostatistics, vol. 9, no. 2, pp. 175–184, 2013.View at: Publisher Site | Google Scholar
Y.-J. Xu and Z.-J. Cai, “Method based on fuzzy linguistic judgement matrix and trapezoidal fuzzy induced ordered weighted geometric (TFIOWG) operator for multi-attribute decision-making problems,” in Proceedings of the International Conference on Wireless Communications, Networking and Mobile Computing (WiCOM '07), vol. 1–15, pp. 5752–5755, September 2007.View at: Publisher Site | Google Scholar
G. W. Wei, R. Lin, X. F. Zhao, and H. J. Wang, “TOPSIS-based linear-programming methodology for multiple attribute decision making with incomplete weight information in intuitionistic fuzzy setting,” Information, vol. 13, no. 5, pp. 1721–1726, 2010.View at: Google Scholar
Y. Xu and Z. Cai, “Standard deviation method for determining the weights of group multiple attribute decision making under uncertain linguistic environment,” in Proceedings of the 7th World Congress on Intelligent Control and Automation (WCICA '08), pp. 8307–8310, June 2008.View at: Publisher Site | Google Scholar