1 Introduction
Analytic hierarchy process (AHP) is a multicriteria decisionmaking method that helps the decisionmakers facing a complex problem with multiple conflicting and subjective criteriaishizaka2009analytic . AHP has been widely applied in various areaszhu2016hesitant ; lin2015measuring ; yepes2015cognitive ; wang2011linear . Based on a hierarchical structure, the criteria weights can be derived from the pairwise comparison matrix. Pairwise comparison matrix as a crucial component of AHP, is commonly utilized to estimate preference values of finite alternatives with respect to a given criterion.
However, the pairwise comparison matrix is difficult to complete in the following cases, which obstructs the subsequent operations of the classical AHP.

The experts lack the knowledge of one or more alternatives.

Partial data in pairwise comparison matrix has been lost.

Huge number of pairwise comparison requested: for n alternatives. For example, 8 alternatives and 6 criteria require 183 entriesishizaka2011review .
Hitherto, it has attracted widespread attention over the issue of incomplete pairwise comparison matrix in AHP. There are quantities of methods having been developed to address the incomplete comparison matrices. Some of them use the incomplete matrix to assign priority weight to each alternative without completionharker1987incomplete ; carmone1997monte ; hua2008ds ; xu2013logarithmic ; dopazo2011parametric . Some methods are devoted to estimate the missing values in matrixfedrizzi2007incomplete ; ergu2016estimating ; bozoki2010optimal ; chen2015bridging ; zhang2012linear ; hu2006backpropagation ; alonso2008consistency ; benitez2015consistent ; gomez2010estimation ; buyukozkan2012new ; cabrerizo2010managing . While these methods have a good effect on allocating the priority weights to alternatives, the methods that focus on matrix completion are more “wellrounded” to restore the preference relations between each pair of alternatives, which are inaccessible before completion.
After reviewing the existing completion methods for incomplete pairwise comparison matrix in AHP, it can be found that the vast majority of these methods focus on the optimal consistency of pairwise comparison matrixfedrizzi2007incomplete ; ergu2016estimating ; bozoki2010optimal ; chen2015bridging ; zhang2012linear . For example, Fedrizzi and Giovefedrizzi2007incomplete proposed a new method to complete the pairwise comparison matrix by minimizing its global inconsistency. Ergu et al.ergu2016estimating
extended the geometric mean induced bias matrix to estimate the missing judgments and improve the consistency ratios by the least absolute error method and the least square method. Based on the graph theory, Bozoíki
et al.bozoki2010optimal developed a completion method which focuses on the optimal consistency and estimates the missing values through iteration. Besides, Hu and Tsaihu2006backpropagation proposed a wellknown back propagation multilayer perception to estimate the missing comparisons of incomplete pairwise comparison matrices in AHP. Alonso et al.alonso2008consistency defined a way to calculate the consistency of incomplete pairwise comparison matrix and maintain its consistency level after completion. Benítez et al.benitez2015consistent used the Frobenius norm to obtain the most consistent/similar complete matrix with the initial incomplete matrix. Gomez et al.gomez2010estimationproposed a completion method based on the neural network. These methods are useful but have some limitations:

Most methods complete the pairwise comparison matrix with optimal consistency, which sometimes can not restore the incomplete matrix to its original state, especially when the initial matrix is not high consistent.

The methods are computationally expensive.
Taking into account these issues, a DEMATELbased completion method for incomplete pairwise comparison matrix in AHP is proposed. Based on DEMATEL method which has ability to derive the total relation matrix from direct relation matrix, the completion of incomplete pairwise comparison matrix can be divided into three steps.

Firstly, turn the incomplete comparison matrix into direct relation matrix.

Secondly, DEMATEL method is adopted to obtain the total relation matrix.

Thirdly, transform the total relation matrix into pairwise comparison matrix with reciprocal preference relations.
The proposed method provides a new perspective to estimate the missing values in pairwise comparison matrix with explicit physical meaning (i.e. by calculating the total relations among alternatives). Besides, the proposed method has low computational cost.
The rest of this paper is organized as follows. Section 2 introduces some preliminaries of this work. Section 3 illustrates the procedure of the proposed method and an application for ranking tennis players is presented. Section 4 discusses the superiority and efficiency of the proposed method and section 5 ends the paper with the conclusion.
2 Preliminaries
2.1 Analytic Hierarchy Process (AHP)
AHP is a multicriteria decisionmaking method developed by Saatysaaty1980analytic and aims at quantifying relative weights for a given set of criteria on a ratio scalexu2015deviation . As a decisionmaking approach, it permits a hierarchical structure of criteria, which provides users with a better focus on specific criteria and subcriteria when allocating the weightsishizaka2013multi . Besides, AHP has ability to judge the consistency of pairwise comparison matrix to show its potential conflicts in decisionmaking process. AHP has been widely used in supply chain managementchan2007global ; deng2014supplier ; chan2006ahp , human reliability analysissu2015dependence , cloud computingliu2016decision , energy planningishizaka2016energy ; onar2015multi ; kahraman2009comparative , site selectionerdougan2016combined ; lee2015integrated ; pourahmad2015combination , software assessmenthu2013software ; hu2015cost and performance evaluationzavadskas2015selecting ; wu2010evaluating . AHP can be extended further by other theories and methods such as fuzzy numberchan2008global ; wang2008extent , choquet integralcorrente2015combining
, kmeans algorithm
lolli2014new , VIKORbuyukozkan2015evaluation and TOPSISerdogan2014evaluating ; kahraman2015fuzzy .Generally, the process of applying AHP is divided into three steps. Firstly, establish a hierarchical structure by recursively decomposing the decision problem. Secondly, construct the pairwise comparison matrix to indicate the relative importance of alternatives. A numerical rating including nine rank scales is suggested, as shown in Tab.1. Thirdly, verify the consistency of pairwise comparison matrix and calculate the priority weights of alternatives. For the completeness of explanation, several basic concepts and formulas are introduced as follows.
Scale  Meaning 

1  Equal importance 
3  Moderate importance 
5  Strong importance 
7  Demonstrated importance 
9  Extreme importance 
2, 4, 6, 8  Intermediate values 
Definition 1
(Pairwise comparison matrix) Assume are alternatives available for decisionmaking, the pairwise comparison matrix is indicated as , and satisfies:
(1) 
where represents the relative importance of over .
Consistency checking is introduced in AHP to verify the usability of constructed pairwise comparison matrix.
Definition 2
(Consistency ratio) For the pairwise comparison matrix , let
denote the largest eigenvalue of
, consistency index () is defined as(2) 
Based on , consistency ratio (CR) is defined as
(3) 
where is the random consistency index related to the dimension of matrix, listed in Tab.2.
n  1  2  3  4  5  6  7  8  9  10 

0  0  0.52  0.89  1.12  1.26  1.36  1.41  1.46  1.49 
If
, the constructed pairwise comparison matrix is considered acceptable and the priority weights of alternatives can be obtained through the eigenvector as Def.
3. Otherwise, the comparison matrix needs to be reconstructed.Definition 3
(Eigenvector) For the pairwise comparison matrix with acceptable consistency, suppose is the eigenvector of , whose is indicated as the priority weight of the th alternative and calculated by
(4) 
2.2 DEMATEL method
Decisionmaking trial and evaluation laboratory (DEMATEL) method was originally developed by Battelle Memorial Institute of Geneva Research Center, aimed at the fragmented and antagonistic phenomena of world societies and searched for integrated solutiongabus1972world ; fontela1976dematel . DEMATEL is built on the basis of graph theory, effectively assessing the total relations and constructing a map between system components in respect to their type and severitytzeng2007evaluating . Reviewing the former studies, DEMATEL has been successfully applied in many diverse areas such as service quality analysisshieh2010dematel ; tseng2009causal , stock selectionshen2015combined and supply chain managementISI:000368039100008 ; wu2015case ; wu2015exploring . DEMATEL can be further extended by other theories and methods such as fuzzy numbersmentes2015fsa ; tsai2015using and ANPISI:000356990800033 ; tzeng2012combined ; ISI:000262178000049 ; wu2008choosing .
The procedure of DEMATEL is divided into five steps.
Step 1: Define quality feature and establish measurement scale
Quality feature is a set of influential characteristics that impact the sophisticated system, which can be determined by literature review, brainstorming, or expert evaluation. After defining the influential characteristics in researching system, establish the measurement scale for the causal relationship and pairwise comparison among influential characteristics. Four levels 0,1,2,3 are suggested, respectively meaning “no impact”, “low impact”, “high impact” and “extreme high impact”. In this step, factors and their direct relations are displayed by a weighted and directed graph.
Step 2: Extract the direct relation matrix of influential factors
In this step, transformation from the weighted and directed graph into direct relation matrix has been done. For influential characteristics , direct relation matrix is denoted as , where is the direct relation of over based on the measurement scale and satisfies if .
Step 3: Normalize the direct relation matrix
The normalized direct relations of factors are a mapping from to , defined as below.
Definition 4
(Normalized direct relation matrix) For the framework of influential characteristics , normalized matrix of direct relation matrix is obtained by
(5) 
Note that there must be at least one row of matrix satisfying lin2008causal . Hence, based on the studies of papoulis1985probability and goodman1988introduction
, substochastic matrix can be obtained through utilizing the normalized direct relation matrix
and absorbing state of Markov chain matrices, shown in (
6).(6) 
where is the null matrix and
is the identity matrix.
Step 4: Calculate the total relation matrix
Due to the characteristic of normalized direct relation matrix , total relation matrix which contains direct and indirect relations among factors can be derived from (7).
Definition 5
(Total relation matrix) For the framework of influential characteristics , assume is the normalized direct relation matrix, total relation matrix is defined as
(7) 
where is the identity matrix.
Step 5: Construct the DEMATEL map
Based on the sum of each row and column of the total relation matrix , and can be obtained. is defined as the prominence, indicating the importance of the ith influential factor. classifies the ith influential factor into the cause or effect category in researching system. If , the factor is regarded as a cause factor. If , the factor is an effect one. The DEMATEL map is shown as Fig.1.
3 DEMATELbased completion method
In this section, a DEMATELbased completion method for incomplete pairwise comparison matrix in AHP is proposed. Moreover, an application for ranking top tennis players is presented.
3.1 Procedure of DEMATELbased completion method
In this subsection, the procedure of DEMATELbased completion method is presented and illustrated by an example. Fig.2 shows the process of the proposed method.
Assume an incomplete pairwise comparison matrix and the unavailable/missing values are displayed by ‘*’.
Step 1: From incomplete pairwise comparison matrix to direct relation matrix.
Pairwise comparison matrix reflects the preference relations between each pair of factors, where in indicates the relative importance of factor to factor . Hence, the known values in can be put into the direct relation matrix without any transformation and the unavailable ones(‘*’) are substituted by 0. The direct relation matrix is as:
Step 2: From direct relation matrix to total relation matrix.
First of all, normalize the direct relation matrix according to (5). Then, based on the normalized matrix and (7), total relation matrix can be obtained. The normalized direct relation matrix and total relation matrix are calculated as:
Step 3: From total relation matrix to pairwise comparison matrix.
In fact, the total relation matrix has completed the unavailable values in initial incomplete matrix. Nevertheless, in pairwise comparison matrix, the multiplication of each pair of symmetric values along the diagonal line is required to equal 1, yet the values in total relation matrix are between 0 and 1. Based on (8), it is feasible to accomplish this transformation from the total relation matrix to pairwise comparison matrix .
(8) 
Hence, the pairwise comparison matrix transformed from is:
Finally, based on the proposed method, the completion for incomplete pairwise comparison matrix has been accomplished as:
3.2 An application for ranking tennis players
Professional tennis as a global popular sport owns large numbers of viewers around the world.
The professional tennis associations such as ATP have been collecting data about the tournaments and the players.
There is a free available database that collects the results of the top tennis players since 1973 (see at http://www.atpw
orldtour.com/Players/HeadToHead.aspx).
In original statistics, 25 tennis players who have been the champion on ATP since 1973 are selected. The statistical competition results matrix is shown in Tab.3. However, the statistical matrix is not complete since some of players did not have the chance to compete with each other or the partial data has been lost. Thus, DEMATELbased completion method is adopted to rank the tennis players based on these statistics. The process is divided into three steps, which is shown in Fig.3.
Aga  Bec  Bor  Con  Cou  Djo  Edb  Fed  Fer  Hew  Kaf  Kue  Len  McE  Moy  Mus  Nad  Nas  New  Raf  Rio  Rod  Saf  Sam  Wil  
Agassi  10/14  2/2  5/12  6/9  3/11  2/5  4/8  8/12  7/11  2/8  2/4  3/4  5/9  0/2  10/15  1/3  5/6  3/6  14/34  5/7  
Becker  4/14  6/6  6/7  25/35  1/1  4/6  10/21  8/10  2/4  2/3  1/1  2/3  3/5  0/1  7/19  7/10  
Borg  15/23  6/8  7/14  10/15  1/4  1/1  
Connors  0/2  0/6  8/23  0/3  6/12  13/34  14/34  12/27  2/4  0/2  0/5  
Courier  7/12  1/7  3/3  6/10  1/6  1/1  0/4  2/3  2/3  7/12  0/1  0/3  0/3  1/2  4/20  
Djokovic  15/31  2/3  6/7  2/4  17/39  4/9  0/2  
Edberg  3/9  10/35  6/12  4/10  1/3  14/27  6/13  1/1  10/10  3/3  1/1  6/14  9/20  
Federer  8/11  16/31  10/13  18/26  2/6  1/3  7/7  10/32  0/3  2/2  21/24  10/12  1/1  
Ferrero  3/5  1/3  3/13  4/10  1/3  3/5  8/14  2/9  2/3  3/4  0/5  6/12  
Hewitt  4/8  0/1  1/7  8/26  6/10  7/8  3/4  7/12  4/10  3/4  3/5  7/14  7/14  5/9  
Kafelnikov  4/12  2/6  5/6  2/3  4/6  2/3  1/8  5/12  3/6  1/5  3/5  6/8  2/4  2/13  1/2  
Kuerten  4/11  0/1  2/3  2/5  1/4  7/12  4/7  3/3  4/8  2/4  1/2  4/7  1/3  
Lendl  6/8  11/21  2/8  21/34  4/4  13/27  21/36  4/5  1/1  0/1  3/8  15/22  
McEnroe  2/4  2/10  7/14  20/34  1/3  7/13  15/36  6/9  1/2  0/3  7/13  
Moya  1/4  2/4  1/3  2/4  0/1  0/7  6/14  5/12  3/6  3/7  4/8  2/8  3/4  2/7  1/5  4/7  1/4  
Muster  4/9  1/3  5/12  0/10  4/5  0/3  1/5  4/8  0/3  3/4  0/1  2/11  0/2  
Nad  2/2  22/39  22/32  7/9  6/10  6/8  7/10  2/2  
Nastase  0/1  5/15  15/27  1/1  0/1  3/9  4/5  0/1  
Newcombe  3/4  2/4  1/2  1/5  
Rafter  5/15  1/3  3/3  0/3  3/3  1/3  1/4  2/5  4/8  1/1  1/4  3/3  2/3  1/1  4/16  1/3  
Rios  2/3  2/5  3/3  0/1  0/2  1/4  2/5  2/8  2/4  5/7  1/4  1/3  0/2  1/4  0/2  
Roddick  1/6  5/9  3/24  5/5  7/14  1/2  4/5  3/10  2/2  4/7  2/3  
Safin  3/6  1/1  1/2  2/2  2/12  6/12  7/14  2/4  3/7  3/7  1/1  0/2  0/1  3/4  3/7  4/7  
Sampras  20/34  12/19  2/2  16/20  8/14  0/1  4/9  11/13  2/3  5/8  3/3  3/4  9/11  12/16  2/2  1/3  3/7  2/3  
Wilander  2/7  3/10  0/1  5/5  11/20  1/2  7/22  6/13  2/2  1/1  2/3  1/3  
Note: Due to the space limitation, the names of tennis players have been abbreviated to the first three letters in the horizontal row and their full names can be seen in the vertical column. 
Aga  Bec  Bor  Con  Cou  Djo  Edb  Fed  Fer  Hew  Kaf  Kue  Len  McE  Moy  Mus  Nad  Nas  New  Raf  Rio  Rod  Saf  Sam  Wil  
Agassi  1.39  1.07  0.90  1.17  0.76  0.95  1.00  1.24  1.17  0.80  1.00  1.12  1.05  0.93  1.31  0.95  1.28  1.00  0.73  1.18  
Becker  0.72  1.38  1.38  2.28  1.04  1.18  0.95  1.43  1.00  1.05  1.03  1.05  1.05  0.97  0.77  1.24  
Borg  1.45  1.25  1.00  1.31  0.89  1.03  
Connors  0.93  0.73  0.69  0.88  1.00  0.66  0.73  0.86  1.00  0.93  0.78  
Courier  1.11  0.72  1.13  1.11  0.78  1.03  0.83  1.05  1.05  1.11  0.97  0.88  0.88  1.00  0.49  
Djokovic  0.95  1.06  1.38  1.00  0.77  0.95  0.93  
Edberg  0.85  0.44  1.00  0.90  0.95  1.05  0.95  1.03  1.89  1.13  1.03  0.90  0.90  
Federer  1.32  1.05  1.49  1.72  0.90  0.95  1.48  0.52  0.88  1.07  3.31  1.64  1.03  
Ferrero  1.05  0.95  0.67  0.90  0.95  1.05  1.11  0.75  1.05  1.12  0.78  1.00  0  
Hewitt  1.00  0.97  0.72  0.58  1.11  1.49  1.12  1.11  0.90  1.12  1.05  1.00  1.00  1.05  
Kafelnikov  0.81  0.90  1.28  1.05  1.11  1.05  0.67  0.90  1.00  0.84  1.05  1.25  1.00  0.57  1.00  
Kuerten  0.85  0.97  1.05  0.95  0.89  1.11  1.05  1.13  1.00  1.00  1.00  1.05  0.95  
Lendl  1.25  1.05  0.80  1.52  1.20  0.95  1.36  1.19  1.03  0.97  0.90  1.54  
McEnroe  1.00  0.70  1.00  1.36  0.95  1.05  0.73  1.17  1.00  0.88  1.05  
Moya  0.89  1.00  0.95  1.00  0.97  0.67  0.90  0.90  1.00  0.95  1.00  0.80  1.12  0.85  0.84  1.05  0.89  
Muster  0.95  0.95  0.90  0.53  1.19  0.88  0.84  1.00  0.88  1.12  0.97  0.65  0.93  
Nadal  1.07  1.29  1.91  1.34  1.11  1.25  1.24  1.07  
Nastase  0.97  0.77  1.17  1.03  0.96  0.85  1.19  0.97  
Newcombe  1.12  1.00  1.00  0.84  
Rafter  0.77  0.95  1.13  0.88  1.13  0.95  0.89  0.95  1.00  1.04  0.89  1.13  1.05  1.03  0.64  0.95  
Rios  1.05  0.95  1.13  0.97  0.93  0.89  0.95  0.80  1.00  1.18  0.89  0.95  0.93  0.89  0.93  
Roddick  0.78  1.05  0.30  1.28  1.00  1.00  1.19  0.80  1.07  1.05  1.05  
Safin  1.00  1.03  1.00  1.07  0.61  1.00  1.00  1.00  0.95  0.95  1.03  0.93  0.97  1.12  0.95  1.05  
Sampras  1.36  1.30  1.07  2.04  1.11  0.97  0.95  1.77  1.05  1.11  1.13  1.12  1.53  1.57  1.07  0.95  0.95  1.05  
Wilander  0.85  0.80  0.97  1.28  1.11  1.00  0.65  0.95  1.07  1.03  1.05  0.95  
Note: Due to the space limitation, the names of tennis players have been abbreviated to the first three letters in the horizontal row and their full names can be seen in the vertical column. 
Aga  Bec  Bor  Con  Cou  Djo  Edb  Fed  Fer  Hew  Kaf  Kue  Len  McE  Moy  Mus  Nad  Nas  New  Raf  Rio  Rod  Saf  Sam  Wil  
Agassi  1.00  1.39  0.89  1.07  0.90  1.00  1.17  0.76  0.95  1.00  1.24  1.17  0.80  1.00  1.12  1.05  0.93  1.06  1.08  1.31  0.95  1.28  1.00  0.73  1.18 
Becker  0.72  1.00  0.99  1.38  1.38  0.96  2.28  0.83  1.04  1.04  1.18  1.04  0.95  1.43  1.00  1.05  0.81  1.03  1.26  1.05  1.05  1.04  0.97  0.77  1.24 
Borg  1.13  1.01  1.00  1.45  1.25  1.08  1.22  0.94  1.15  1.06  1.19  1.16  1.25  1.00  1.20  1.30  0.92  1.31  0.89  1.19  1.17  1.19  1.13  0.99  1.03 
Connors  0.93  0.73  0.69  1.00  0.88  0.84  1.00  0.72  0.89  0.82  0.94  0.90  0.66  0.73  0.92  1.00  0.72  0.86  1.00  0.92  0.89  0.93  0.86  0.93  0.78 
Courier  1.11  0.72  0.80  1.13  1.00  0.89  1.11  0.74  0.92  0.83  0.78  1.03  0.83  1.05  1.05  1.11  0.75  0.97  1.04  0.88  0.88  0.92  1.00  0.49  0.94 
Djokovic  1.00  1.04  0.92  1.20  1.12  1.00  1.10  0.95  1.05  1.38  1.11  1.04  0.95  1.11  1.00  1.12  0.77  1.09  1.14  1.09  1.08  0.95  0.93  0.93  1.10 
Edberg  0.85  0.44  0.82  1.00  0.90  0.91  1.00  0.79  0.96  0.86  0.95  0.97  1.05  0.95  1.03  1.89  0.75  0.91  0.97  1.13  1.03  0.97  0.94  0.90  0.90 
Federer  1.32  1.21  1.06  1.39  1.36  1.05  1.26  1.00  1.49  1.72  0.90  0.95  1.10  1.29  1.48  1.32  0.52  1.27  1.32  0.88  1.07  3.31  1.64  1.03  1.23 
Ferrero  1.05  0.96  0.87  1.12  1.09  0.95  1.04  0.67  1.00  0.90  0.95  1.05  0.90  1.03  1.11  1.07  0.75  1.03  1.07  1.05  1.12  0.78  1.00  0.82  1.03 
Hewitt  1.00  0.97  0.95  1.22  1.20  0.72  1.17  0.58  1.11  1.00  1.49  1.12  0.97  1.14  1.11  1.17  0.90  1.10  1.17  1.12  1.05  1.00  1.00  1.05  1.13 
Kafelnikov  0.81  0.90  0.84  1.07  1.28  0.90  1.05  1.11  1.05  0.67  1.00  0.90  0.83  0.97  1.00  0.84  0.74  0.98  1.02  1.05  1.25  1.00  1.00  0.57  1.00 
Kuerten  0.85  0.96  0.86  1.11  0.97  0.96  1.03  1.05  0.95  0.89  1.11  1.00  0.89  1.03  1.05  1.13  0.78  1.02  1.06  1.00  1.00  1.00  1.05  0.95  1.03 
Lendl  1.25  1.05  0.80  1.52  1.20  1.05  0.95  0.91  1.12  1.03  1.20  1.12  1.00  1.36  1.15  1.19  0.91  1.03  1.18  0.97  1.13  1.15  1.10  0.90  1.54 
McEnroe  1.00  0.70  1.00  1.36  0.95  0.90  1.05  0.78  0.97  0.88  1.03  0.97  0.73  1.00  1.00  1.10  0.78  1.17  1.00  1.01  0.96  1.00  0.93  0.88  1.05 
Moya  0.89  1.00  0.84  1.08  0.95  1.00  0.97  0.67  0.90  0.90  1.00  0.95  0.87  1.00  1.00  1.00  0.80  0.98  1.03  1.12  0.85  0.84  1.05  0.89  0.99 
Muster  0.95  0.95  0.77  1.00  0.90  0.89  0.53  0.76  0.93  0.86  1.19  0.88  0.84  0.91  1.00  1.00  0.74  0.91  0.94  0.88  1.12  0.93  0.97  0.65  0.93 
Nadal  1.07  1.23  1.08  1.39  1.34  1.29  1.33  1.91  1.34  1.11  1.36  1.29  1.10  1.28  1.25  1.35  1.00  1.28  1.33  1.34  1.32  1.24  1.07  1.12  1.32 
Nastase  0.94  0.97  0.77  1.17  1.03  0.92  1.10  0.79  0.97  0.91  1.02  0.98  0.96  0.85  1.02  1.10  0.78  1.00  1.19  1.01  0.99  0.99  0.96  0.80  0.97 
Newcombe  0.93  0.79  1.12  1.00  0.96  0.88  1.03  0.76  0.94  0.86  0.98  0.94  0.85  1.00  0.97  1.06  0.75  0.84  1.00  0.97  0.94  0.96  0.91  0.82  0.96 
Rafter  0.77  0.95  0.84  1.09  1.13  0.92  0.88  1.13  0.95  0.89  0.95  1.00  1.04  0.99  0.89  1.13  0.74  0.99  1.03  1.00  1.05  1.02  1.03  0.64  0.95 
Rios  1.05  0.95  0.85  1.12  1.13  0.92  0.97  0.93  0.89  0.95  0.80  1.00  0.89  1.04  1.18  0.89  0.76  1.01  1.07  0.95  1.00  0.93  0.89  0.93  1.00 
Roddick  0.78  0.96  0.84  1.08  1.08  1.05  1.03  0.30  1.28  1.00  1.00  1.00  0.87  1.00  1.19  1.07  0.80  1.01  1.04  0.98  1.07  1.00  1.05  1.05  1.00 
Safin  1.00  1.03  0.88  1.16  1.00  1.07  1.06  0.61  1.00  1.00  1.00  0.95  0.91  1.07  0.95  1.03  0.93  1.04  1.10  0.97  1.12  0.95  1.00  1.05  1.05 
Sampras  1.36  1.30  1.01  1.07  2.04  1.07  1.11  0.97  1.22  0.95  1.77  1.05  1.11  1.13  1.12  1.53  0.89  1.25  1.22  1.57  1.07  0.95  0.95  1.00  1.05 
Wilander  0.85  0.80  0.97  1.28  1.06  0.91  1.11  0.81  0.97  0.88  1.00  0.97  0.65  0.95  1.01  1.07  0.76  1.03  1.05  1.05  1.00  1.00  0.96  0.95  1.00 
Note: Due to the space limitation, the names of tennis players have been abbreviated to the first three letters in the horizontal row and their full names can be seen in the vertical column. 
Firstly, the statistical competition results matrix (i.e. with additive preference relations) is converted into the pairwise comparison matrix (i.e. with multiplicative preference relations) following the method described in bozoki2016application . Fig.4 illustrates this transformation process. The converted pairwise comparison matrix is shown in Tab.4.
Secondly, DEMATELbased completion method is adopted to complete the pairwise comparison matrix. Following the procedure of the proposed method, the complete pairwise comparison matrix is obtained, which is shown in Tab.5.
Thirdly, based on the complete pairwise comparison matrix, the priorities of tennis players can be calculated via (4). Tab.6 shows the final ranking of the 25 tennis players.
Rank  Player  Priority  Rank  Player  Priority 

1  Nadal  0.0503  14  Rios  0.0380 
2  Federer  0.0503  15  McEnroe  0.0380 
3  Sampras  0.0467  16  Nastase  0.0380 
4  Borg  0.0444  17  Rafter  0.0378 
5  Lendl  0.0436  18  Wilander  0.0378 
6  Becker  0.0430  19  Kafelnikov  0.0375 
7  Hewitt  0.0414  20  Edberg  0.0374 
8  Djokovic  0.0412  21  Moya  0.0371 
9  Agassi  0.0409  22  Newcombe  0.0365 
10  Safin  0.0392  23  Courier  0.0360 
11  Kuerten  0.0390  24  Muster  0.0353 
12  Roddick  0.0384  25  Connors  0.0339 
13  Ferrero  0.0383 
4 Discussion
The result in last section indicates that the proposed method has ability to complete the pairwise comparison matrices in AHP. In this section, we would further analyze the superiority and efficiency of the proposed method.
The proposed method has the following advantages:

The proposed method estimates the missing values in pairwise comparison matrix from a new perspective (i.e. by calculating the total relations among alternatives).

The proposed method has low computational cost.
After reviewing the prior studies, an assessment for the computational cost of the existing completion methods and the proposed method is done. The methods for comparison satisfy:

The methods belong to completion methods, which have ability to estimate the missing values in pairwise comparison matrix.

The pairwise comparison matrices, which can be completed by these completion methods, present reciprocal preference relations among alternatives. In other words, the matrices request that the multiplication of each pair of symmetric values along the diagonal line is equal to 1.
As for these completion methods, some of them estimate the missing values through iteration, or based on the complex theories/methods, which causes the methods expensive computational cost. For example, Bozoíki et al.bozoki2010optimal developed a completion method which estimates the missing values in pairwise comparison matrix through iteration and optimizes the matrix consistency based on graph theory. Gomez et al.gomez2010estimation
proposed a completion method based on the neural network. Neural network as one of the branches of artificial intelligence leads to the complexity of the completion method. As for the completion methods without iteration or complex theories/algorithms such as the methods listed in Tab.
7, time complexity is used to accurately reflect their computational cost. Assume there is an incomplete pairwise comparison matrix, which has n missing values, the time complexity of these methods is shown in Tab.8.Description  Source  

Method 1 

Ergu et al.ergu2016estimating  
Method 2 

Alonso et al.alonso2008consistency  
Method 3 

Benítez et al.benitez2015consistent 
Method 1  Method 2  Method 3  The proposed method  

Time complexity 
It can be seen from the Tab.8 that the time complexity of the proposed method is lower than the others. In addition, the most existing completion methods have the following two operations:

Divide the matrix into upper and lower triangular matrix. The pairwise comparison matrix in AHP has a property that the multiplication of each pair of symmetric values along the diagonal line is equal to 1. These methods firstly focus on the completion of upper/lower triangle of incomplete matrix and use the property to accomplish the other triangular matrix.

Traverse the pairwise comparison matrix to obtain the positions of unavailable values, then estimate these missing values in sequence.
However, the proposed method completes the matrix from a holistic viewpoint without dividing the matrix into the upper and lower triangular matrix. Furthermore, based on DEMATEL, all the missing values in matrix can be simultaneously estimated. Hence, following the procedure of the proposed method, the completion of incomplete pairwise comparison matrix becomes straightforward and efficient.
5 Conclusion
Pairwise comparison matrix plays a pivotal role in AHP. However, in many cases, only partial information in pairwise comparison matrix is available, which obstructs the subsequent operations of the classical AHP. In this paper, we propose a new completion method for incomplete pairwise comparison matrix in AHP. Based on DEMATEL, the proposed method has ability to complete the matrix by obtaining the total (direct and indirect) relations from the direct preference relations between each pair of alternatives. Furthermore, an application of the proposed method for ranking tennis players is presented. From the ranking result and the comparison with several representative completion methods, the proposed method is proved to be effective and efficient. The proposed method provides a new perspective to complete the matrix with explicit physical meaning. Besides, the proposed method has low computation cost. This promising method has a wide application in multicriteria decisionmaking. In our further study, we would extend the proposed method for the incomplete fuzzy pairwise comparison matrices.
Acknowledgment
The work is partially supported by National Natural Science Foundation of China (Grant Nos. 61573290,61503237), China State Key Laboratory of Virtual Reality Technology and Systems, Beihang University (Grant No.BUAAVR14KF02).
References
References
 (1) A. Ishizaka, A. Labib, Analytic hierarchy process and expert choice: Benefits and limitations, OR Insight 22 (4) (2009) 201–220.
 (2) B. Zhu, Z. Xu, R. Zhang, M. Hong, Hesitant analytic hierarchy process, European Journal of Operational Research 250 (2) (2016) 602–614.
 (3) C. T. Lin, Y. H. Goay, Measuring National Tourism Organization Abroad Office Competitiveness, International Journal of Tourism Research 17 (2) (2015) 118–129.
 (4) V. Yepes, T. GarcíaSegura, J. MorenoJiménez, A cognitive approach for the multiobjective optimization of RC structural problems, Archives of Civil and Mechanical Engineering 15 (4) (2015) 1024–1036.
 (5) Y.M. Wang, K.S. Chin, A linear goal programming approach to determining the relative importance weights of customer requirements in quality function deployment, Information Sciences 181 (24) (2011) 5523–5533.
 (6) A. Ishizaka, A. Labib, Review of the main developments in the analytic hierarchy process, Expert systems with applications 38 (11) (2011) 14336–14345.
 (7) P. T. Harker, Incomplete pairwise comparisons in the analytic hierarchy process, Mathematical Modelling 9 (11) (1987) 837–848.
 (8) F. J. Carmone, A. Kara, S. H. Zanakis, A Monte Carlo investigation of incomplete pairwise comparison matrices in AHP, European Journal of Operational Research 102 (3) (1997) 538–553.
 (9) Z. Hua, B. Gong, X. Xu, A DS–AHP approach for multiattribute decision making problem with incomplete information, Expert systems with applications 34 (3) (2008) 2221–2227.
 (10) Y. Xu, R. Patnayakuni, H. Wang, Logarithmic least squares method to priority for group decision making with incomplete fuzzy preference relations, Applied Mathematical Modelling 37 (4) (2013) 2139–2152.
 (11) E. Dopazo, M. RuizTagle, A parametric GP model dealing with incomplete information for group decisionmaking, Applied Mathematics and Computation 218 (2) (2011) 514–519.
 (12) M. Fedrizzi, S. Giove, Incomplete pairwise comparison and consistency optimization, European Journal of Operational Research 183 (1) (2007) 303–313.
 (13) D. Ergu, G. Kou, Y. Peng, M. Zhang, Estimating the missing values for the incomplete decision matrix and consistency optimization in emergency management, Applied Mathematical Modelling 40 (1) (2016) 254–267.
 (14) S. Bozóki, J. Fülöp, L. Rónyai, On optimal completion of incomplete pairwise comparison matrices, Mathematical and Computer Modelling 52 (1) (2010) 318–333.
 (15) K. Chen, G. Kou, J. M. Tarn, Y. Song, Bridging the gap between missing and inconsistent values in eliciting preference from pairwise comparison matrices, Annals of Operations Research 235 (1) (2015) 155–175.
 (16) G. Zhang, Y. Dong, Y. Xu, Linear optimization modeling of consistency issues in group decision making based on fuzzy preference relations, Expert Systems with Applications 39 (3) (2012) 2415–2420.

(17)
Y.C. Hu, J.F. Tsai, Backpropagation multilayer perceptron for incomplete pairwise comparison matrices in analytic hierarchy process, Applied Mathematics and Computation 180 (1) (2006) 53–62.
 (18) S. Alonso, F. Chiclana, F. Herrera, E. HerreraViedma, J. AlcaláFdez, C. Porcel, A consistencybased procedure to estimate missing pairwise preference values, International Journal of Intelligent Systems 23 (2) (2008) 155–175.
 (19) J. Benítez, X. DelgadoGalván, J. Izquierdo, R. PérezGarcía, Consistent completion of incomplete judgments in decision making using AHP, Journal of Computational and Applied Mathematics 290 (2015) 412–422.
 (20) J. A. GomezRuiz, M. Karanik, J. I. Peláez, Estimation of missing judgments in AHP pairwise matrices using a neural networkbased model, Applied Mathematics and Computation 216 (10) (2010) 2959–2975.
 (21) G. Büyüközkan, G. Çifçi, A new incomplete preference relations based approach to quality function deployment, Information Sciences 206 (2012) 30–41.
 (22) F. J. Cabrerizo, I. J. Pérez, E. HerreraViedma, Managing the consensus in group decision making in an unbalanced fuzzy linguistic context with incomplete information, KnowledgeBased Systems 23 (2) (2010) 169–181.
 (23) T. L. Saaty, The analytic hierarchy process: planning, priority setting, resources allocation, New York: McGraw, 1980.
 (24) Z. Xu, Deviation square priority method for distinct preference structures based on generalized multiplicative consistency, Fuzzy Systems, IEEE Transactions on 23 (4) (2015) 1164–1180.
 (25) A. Ishizaka, P. Nemery, Multicriteria decision analysis: methods and software, John Wiley & Sons, 2013.
 (26) F. T. Chan, N. Kumar, Global supplier development considering risk factors using fuzzy extended AHPbased approach, Omega 35 (4) (2007) 417–431.
 (27) X. Deng, Y. Hu, Y. Deng, S. Mahadevan, Supplier selection using AHP methodology extended by D numbers, Expert Systems with Applications 41 (1) (2014) 156–167.
 (28) F. T. Chan, H. Chan, H. C. Lau, R. W. Ip, An AHP approach in benchmarking logistics performance of the postal industry, Benchmarking: An International Journal 13 (6) (2006) 636–661.
 (29) X. Su, S. Mahadevan, P. Xu, Y. Deng, Dependence assessment in human reliability analysis using evidence theory and AHP, Risk Analysis 35 (7) (2015) 1296–1316.
 (30) S. Liu, F. T. Chan, W. Ran, Decision making for the selection of cloud vendor: An improved approach under group decisionmaking with integrated weights and objective/subjective attributes, Expert Systems with Applications 55 (2016) 37–47.
 (31) A. Ishizaka, S. Siraj, P. Nemery, Which energy mix for the UK (United Kingdom)? An evolutive descriptive mapping with the integrated GAIA (graphical analysis for interactive aid)–AHP (analytic hierarchy process) visualization tool, Energy 95 (2016) 602–611.
 (32) S. C. Onar, B. Oztaysi, İ. Otay, C. Kahraman, Multiexpert wind energy technology selection using intervalvalued intuitionistic fuzzy sets, Energy 90 (1) (2015) 274–285.
 (33) C. Kahraman, İ. Kaya, S. Cebi, A comparative analysis for multiattribute selection among renewable energy alternatives using fuzzy axiomatic design and fuzzy analytic hierarchy process, Energy 34 (10) (2009) 1603–1616.
 (34) M. Erdoğan, İ. Kaya, A combined fuzzy approach to determine the best region for a nuclear power plant in Turkey, Applied Soft Computing 39 (2016) 84–93.
 (35) A. H. Lee, H.Y. Kang, C.Y. Lin, K.C. Shen, An Integrated DecisionMaking Model for the Location of a PV Solar Plant, Sustainability 7 (10) (2015) 13522–13541.
 (36) A. Pourahmad, A. Hosseini, A. Banaitis, H. Nasiri, N. Banaitienė, G.H. Tzeng, Combination of fuzzyAHP and DEMATELANP with gis in a new hybrid mcdm model used for the selection of the best space for leisure in a blighted urban site, Technological and Economic Development of Economy 21 (5) (2015) 773–796.

(37)
Y. Hu, X. Zhang, E. Ngai, R. Cai, M. Liu, Software project risk analysis using bayesian networks with causality constraints, Decision Support Systems 56 (2013) 439–449.
 (38) Y. Hu, B. Feng, X. Mo, X. Zhang, E. Ngai, M. Fan, M. Liu, Costsensitive and ensemblebased prediction model for outsourced software project risk prediction, Decision Support Systems 72 (2015) 11–23.
 (39) E. K. Zavadskas, Z. Turskis, J. Antucheviciene, Selecting a Contractor by Using a Novel Method for Multiple Attribute Analysis: Weighted Aggregated Sum Product Assessment with Grey Values (WASPASG), Studies in Informatics and Control 24 (2) (2015) 141–150.
 (40) C.R. Wu, C.T. Lin, P.H. Tsai, Evaluating business performance of wealth management banks, European Journal of Operational Research 207 (2) (2010) 971–979.
 (41) F. T. Chan, N. Kumar, M. Tiwari, H. Lau, K. Choy, Global supplier selection: a fuzzyAHP approach, International Journal of Production Research 46 (14) (2008) 3825–3857.
 (42) Y.M. Wang, Y. Luo, Z. Hua, On the extent analysis method for fuzzy AHP and its applications, European Journal of Operational Research 186 (2) (2008) 735–747.
 (43) S. Corrente, S. Greco, A. Ishizaka, Combining analytical hierarchy process and choquet integral within nonadditive robust ordinal regression, Omega 61 (2016) 2–18.
 (44) F. Lolli, A. Ishizaka, R. Gamberini, New AHPbased approaches for multicriteria inventory classification, International Journal of Production Economics 156 (2014) 62–74.
 (45) G. Büyüközkan, A. Görener, Evaluation of product development partners using an integrated AHPVIKOR model, Kybernetes 44 (2) (2015) 220–237.
 (46) M. Erdogan, I. Kaya, Evaluating AlternativeFuel Busses for Public Transportation in Istanbul Using Interval Type2 Fuzzy AHP and TOPSIS, Journal of multiplevalued logic and soft computing 26 (6) (2016) 625–642.
 (47) C. Kahraman, A. Suder, E. T. Bekar, Fuzzy multiattribute consumer choice among health insurance options, Technological and Economic Development of Economy 22 (1) (2016) 1–20.
 (48) A. Gabus, E. Fontela, World problems, an invitation to further thought within the framework of DEMATEL, Battelle Geneva Research Center, Geneva, Switzerland, 1972.
 (49) E. Fontela, A. Gabus, The DEMATEL observer, Battelle Geneva Research Centre, Geneva, 1976.
 (50) G.H. Tzeng, C.H. Chiang, C.W. Li, Evaluating intertwined effects in elearning programs: A novel hybrid MCDM model based on factor analysis and DEMATEL, Expert systems with Applications 32 (4) (2007) 1028–1044.
 (51) J.I. Shieh, H.H. Wu, K.K. Huang, A DEMATEL method in identifying key success factors of hospital service quality, KnowledgeBased Systems 23 (3) (2010) 277–282.
 (52) M.L. Tseng, A causal and effect decision making model of service quality expectation using greyfuzzy DEMATEL approach, Expert Systems with Applications 36 (4) (2009) 7738–7748.
 (53) K.Y. Shen, G.H. Tzeng, Combined soft computing model for value stock selection based on fundamental analysis, Applied Soft Computing 37 (2015) 142–155.
 (54) J. J. H. Liou, J. Tamosaitiene, E. K. Zavadskas, G.H. Tzeng, New hybrid COPRASG MADM Model for improving and selecting suppliers in green supply chain management, International Journal of Production Research 54 (1, SI) (2016) 114–134.
 (55) H.H. Wu, S.Y. Chang, A case study of using DEMATEL method to identify critical factors in green supply chain management, Applied Mathematics and Computation 256 (2015) 394–403.
 (56) K.J. Wu, C.J. Liao, M.L. Tseng, A. S. F. Chiu, Exploring decisive factors in green supply chain practices under uncertainty, International Journal of Production Economics 159 (SI) (2015) 147–157.
 (57) A. Mentes, H. Akyildiz, M. Yetkin, N. Turkoglu, A FSA based fuzzy DEMATEL approach for risk assessment of cargo ships at coasts and open seas of Turkey, Safety Science 79 (2015) 1–10.
 (58) S.B. Tsai, M.F. Chien, Y. Xue, L. Li, X. Jiang, Q. Chen, J. Zhou, L. Wang, Using the Fuzzy DEMATEL to Determine Environmental Performance: A Case of Printed Circuit Board Industry in Taiwan, PloS one 10 (6) (2015) e0129153.
 (59) J. J. H. Liou, Building an effective system for carbon reduction management, Journal of Cleaner Production 103 (2015) 353–361.
 (60) G.H. Tzeng, C.Y. Huang, Combined DEMATEL technique with hybrid MCDM methods for creating the aspired intelligent global manufacturing & logistics systems, Annals of Operations Research 197 (1) (2012) 159–190.
 (61) W.H. Tsai, W.C. Chou, Selecting management systems for sustainable development in SMEs: A novel hybrid model based on DEMATEL, ANP, and ZOGP, Expert Systems with Applications 36 (2) (2009) 1444–1458.
 (62) W.W. Wu, Choosing knowledge management strategies by using a combined ANP and DEMATEL approach, Expert Systems with Applications 35 (3) (2008) 828–835.
 (63) C.J. Lin, W.W. Wu, A causal analytical method for group decisionmaking under fuzzy environment, Expert Systems with Applications 34 (1) (2008) 205–213.

(64)
A. Papoulis, S. U. Pillai, Probability, random variables, and stochastic processes, McGrawHill, 1985.
 (65) R. Goodman, Introduction to stochastic models, Courier Corporation, 1988.
 (66) S. Bozóki, L. Csató, J. Temesi, An application of incomplete pairwise comparison matrices for ranking top tennis players, European Journal of Operational Research 248 (1) (2016) 211–218.
Comments
There are no comments yet.