Graduate Institute of Industrial Management Southern Taiwan University Application of a Fuzzy MCDM Model to the Evaluation and Selection of Smart Phones Advisor:Prof. Chu, Ta-Chung Student: Chen, Chih-kai 1 OUTLINE INTRODUCTION LITERATURE REVIEW FUZZY SET THEORY MODEL ESTABLISHMENT NUMERICAL EXAMPLE CONCLUSIONS 2 INTRODUCTION Background and Motivation Why We Apply FMCDM Research Objectives Research Framework 3 Background and Motivation According to Gartner, International Research and Consulting: 1. In the third quarter of 2010, worldwide mobile phone sales volume of 417 million represented the growth of 35% over the same period in 2009. 2. In addition, smart phone sales volume is more substantial in the third quarter of 2009 which grew 96%. 3. The proportion of mobile phone sales volume has slightly increased to 19.3%. 4 The prices of the mobile phones continue to lower and the market tends to be saturated, the manufacturers cannot get as high gross margin as they did in the past. Therefore, manufacturers have begun to research and produce smart phones. The stereotype about the smart phones: 1. These devices are just for the businessmen. 2. It’s expensive. 3. Interfaces are too incomprehensible. 5 Why We Apply FMCDM Evaluating smart phone many criteria (or factors) need to be considered. evaluating smart phone Quantitative Qualitative Hardware Price Brand Awareness of Smart Phone Operating Systems Number of Applications Security Different decision-makers also have different thoughts about the weight of each criterion. 6 Research Objectives The objectives of this study are listed as follows: 1. Smart phone selection related to literature is investigated. Criteria for selecting smart phones are analyzed. A fuzzy MCDM approach is established. A total relative area for ranking fuzzy numbers is suggested. A numerical example is used to demonstrate the computational process of the proposed model. 2. 3. 4. 5. 7 Research Framework Chapter 1 Introduction Chapter 4 Model Establishment Chapter 2 Literature Review Chapter 5 Numerical Example Chapter 3 Fuzzy Set Theory Chapter 6 Conclusion 8 LITERATURE REVIEW Definition of Mobile Phone What is a System of Smart Phone The overview of the system development of the smart phone Criteria Assessment Related literatures on smart phones Related Works on FMCDM Fuzzy Number Ranking Criteria Assessment 9 Definition of Mobile Phone Feature Phones Smart Phones 10 Feature Phones Feature phone has its own mobile phone manufacturer’s operating system (OS), which has a basic audio and video call beside the additional features (e.g. taking photos, sending text message, listening to music, etc.) but it is not allowed to install or remove software (e.g. remove preset program or install GPS software). However, if the phone supports JAVA and BREW, it is able to install applications. The software being developed through the two systems is not user-friendly. 11 Smart Phones Su (2009) Smart phone is a phone equipped with function of phone and PDA. Yang (2009) Smart phones are developed because industrial technology progresses and consumers require integrating multiple requirements into one device. Hsu (2004) In addition to the original function of voice communications, a mobile phone should also be equipped with an open operating system, and sufficient processing power, allowing users to choose application freely and expand multiple or limitless functions. 12 According to the above-mentioned definitions of smart phones, this study defined “Smart Phone” as follows: 1. Opening source operating system platform. 2. Strong support on the third party's applications, and is allowed to install or remove software freely. 3. A strong hardware performance and faster processing power. 13 What is a System of Smart Phone Alter (2002) Operating System (OS) “The system that controls the execution of all other programs, communication with peripheral devices and use of memory and resources.” Malykhina (2007) OS “the heart of the smart phone, it determines a phone's features, performance, security, and application installation.” 14 The overview development of the smart phone According to Gartner's statistics, the leading market of smart phone operating system: Symbian 29,480.1 3Q10 Market share 36.6% Android 20,500.0 25.5% 1,424.5 3.5% iOS Research In Motion 13,484.4 11,908.3 16.7% 14.8% 7,040.4 8,522.7 17.1% 20.7% Microsoft Windows Mobile 22,247.9 2.8% 3,259.9 7.9% 1,697.1 1,214.8 2.1% 1.5% 1,918.5 612.5 4.7% 1.5% 80,532.6 100.0% 41,093.3 100.0% Smart Phone Operating Systems Linux Other OS Total 3Q10 sales 3Q09 sales 18,314.8 3Q09 Market share 44.6% 15 Related literatures on smart phones Yang (2009), “ The Development Trend Analysis of Smart Phone Industry”, he designs an expert questionnaire and interviews with 7 experts. operating interface entertainment platform mobile business specification of software and hardware. 16 Lin and Ye (2009), “ Operating System Battle in the Ecosystem of Smart phone Industry”, they adopt concept of Food Web to explain ecosystem of various smart phone OS. They found out: device maker third-party application developer are two key sources. 17 Information Week (2007), “Survey of Emphasized Functions of Smart phones towards 325 Experts in Related Field”. The results demonstrated that security and easy integration with PC obtain the first and second place respectively. Gizmodo (2008), “Smart phone OS Comparison Chart”. 17 functions are listed in a table and the presence and absence of each function are compared. 18 Many works investigated smart phone operating system as the research object, however, these works cannot offer consumers a method to evaluate and select smart phones. Most importantly, when consumers choose to buy smart phones, they will usually consider both hardware and operating system. This thesis proposes a fuzzy multiple criteria decision making approach to comprehensively consider criteria in hardware and operating system in order to help consumers evaluate and select smart phones. 19 Related literature on fuzzy MCDM In 1970, Bellman and Zadeh introduced fuzzy set theory to multi-criteria decision making, which involved fuzzy decision analysis concepts and models for solving the problem of uncertainty in decision-making. Since then, the fuzzy multiple criteria decision making has resulted in many researches. Fuzzy numbers can be used to better describe suitability of alternatives versus qualitative criteria under fuzzy multiple criteria decision making environment. 20 Fuzzy MCDM is mainly divided into two categories: 1. fuzzy multiple-objective decision-making: It is mainly used in the "planning; design aspects ". 2. fuzzy multiple criteria (attributes) decision-making: It is mainly used in the "assessment; selection aspects". 21 Chang et al. (2009) applied the fuzzy multi-criteria decision making method in enterprise organization to establish the key influential factors for the success of knowledge management. Chou (2007) used fuzzy multiple criteria decision making method to resolve the selection problem of transshipment container port in marine transportation industry. 22 Fuzzy Number Ranking In fuzzy multiple criteria decision making, the final evaluation values are usually still fuzzy numbers. A ranking method is needed to transform these final fuzzy evaluation values into crisp values for decision making. At present, there are many defuzzification methods which have been investigated for ranking fuzzy numbers. 23 Some methods are briefly introduced as follows: Liou and Wang (1992) introduced a total integral value generated by the left and right integral values of a fuzzy number for ranking fuzzy numbers; Chen and Hwang (1992) proposed ranking fuzzy numbers by preference relations, the average of fuzzy number and degree, fuzzy rating, and linguistic terms; Abbasbandy and Hajjari (2009) presented a new approach for ranking trapezoidal fuzzy numbers based on left and right spreads at some α-levels of trapezoidal fuzzy numbers; Farhadinia (2009) proposed a new approach to rank fuzzy numbers based on the concept of lexicographical ordering in order to provide decision makers algorithm in a simple and efficient way. 24 In this research, we suggest total relative area to rank the final fuzzy numbers, and it is developed based on the concept of Chen’s (1985) maximizing set and minimizing set which is one of the most frequently used methods for the problems under fuzzy MCDM environment. 25 Criteria Assessment Evaluation criteria Nature Literature source Qualitative brand awareness of smart phone C1 Benefit This Study brand awareness of OS C2 Benefit J.D. Power(2010) Designs C3 Benefit J.D. Power(2010) Security C4 Benefit Jakajima (2008) Operability C5 Benefit Jakajima (2008) Entertainment C6 Benefit Su (2009) Execution efficiency C7 Benefit Gizmodo (2009) 26 Quantitative Screen size C8 Benefit This Study Camera resolution C9 Benefit This Study C10 Benefit Lin and Ye (2009) Price of applications C11 Cost Distimo (2010) Price C12 Cost This Study Weight C13 Cost This Study Number of applications 27 FUZZY SET THEORY Fuzzy Sets Fuzzy Numbers α-cut Arithmetic Operations on Fuzzy Numbers Linguistic Values 28 Fuzzy Set The fuzzy set A can be expressed as: A {( x, f A ( x)) | x U } (3.1) where U is the universe of discourse, x is an element in U, A is a fuzzy set in U, f x is the membership function of A at x. The larger f x , the stronger the grade of membership for x in A. A A 29 Fuzzy Numbers A real fuzzy number A is described as any fuzzy subset of the real line R with membership function f A which possesses the following properties (Dubois and Prade, 1978): (a) (b) (c) (d) (e) (f) f A is a continuous mapping from R to [0,1]; f A ( x) 0, x (, a] ; f A is strictly increasing on [a ,b]; f A ( x) 1, x b, c; f A is strictly decreasing on [c ,d]; f A ( x) 0, x [d , ) ; where, A can be denoted as a , b , c , d . 30 The membership function f of the fuzzy number A can also be expressed as: A f AL ( x), a x b, bxc 1, f A ( x) R f A ( x), c x d , 0, otherwise, (3.2) where f AL ( x) and f AR ( x) are left and right membership functions of A,respectively. 31 α-cut The α-cuts of fuzzy number A can be defined as: A x | f A x , 0, 1 (3.3) where A is a non-empty bounded closed interval contained in R and can be denoted by A [ A , A ], where Al and Au are its lower and upper bounds, respectively. l u 32 Arithmetic Operations on Fuzzy Numbers Given fuzzy numbers A and B, A, B R , the α-cuts of A and B are A [ A , A ] and B [B , B ], respectively. By interval arithmetic, some main operations of A and B can be expressed as follows (Kaufmann and Gupta, 1991): l u l u A B [ Al Bl , Au Bu ] (3.4) A B [ Al Bu , Au Bl ] (3.5) A B [ Al Bl , Au Bu ] (3.6) A () B A r Al Au [ , ] Bu Bl (3.7) Al r , Au r , r R (3.8) 33 Linguistic Variable According to Zadeh (1975), the concept of linguistic variable is very useful in dealing with situations which are complex to be reasonably described by conventional quantitative expressions. A1=(0,0,0.2)=Unimportant A2=(0.1,0.3,0.5)= Between Unimportant and Important A3=(0.3,0.5,0.7)=Important A4=(0.5,0.7,0.9)=Very important A5=(0.8,1,1)=Absolutely important Figure 3.1. Linguistic Values and Fuzzy Numbers for Degree of Importance 34 MODEL DEVELOPMENT Average Importance Weights Aggregate Ratings of Alternatives versus Qualitative Criteria Normalize Values of Alternatives versus Quantitative Criteria Aggregate the Ratings and Weights Rank Fuzzy Numbers 35 Model Development Dt decision makers, Dt , t 1,2,..., l Ai candidate of smart phones, Ai , i 1,2,..., m C j selected criteria, C j , j 1,2,..., n In model development process, criteria are categorized into three groups: Benefit qualitative criteria: C j , j 1,..., g Benefit quantitative criteria: C j , j g 1,..., h Cost quantitative criteria: C j , j h 1,..., n 36 Average Importance Weights Assume w jt (a jt , b jt , c jt ) , w jt R , j 1,..., n , t 1,..., l , 1 w j (w j1 w j 2 ... w jl ) l (4.1) 1 l 1 l 1 l where, a j a jt , b j b jt , c j c jt . l t 1 l t 1 l t 1 w jt represents the weight assigned by each decision maker for each criterion. w j represents the average importance weight of each criterion. 37 Aggregate Ratings of Alternatives versus Qualitative Criteria Assume xijt (dijt , eijt , fijt ) , i 1,..., m, j 1,..., g , t 1,..., l 1 xij ( xij1 xij 2 ... xijl ) l (4.2) 1 l 1 l 1 l where dij dijt , eij eijt , fij fijt l t 1 l t 1 l t 1 xijt denotes ratings assigned by each decision maker for each alternative versus each qualitative criterion. xij denotes averaged ratings of each alternative versus each qualitative criterion. 38 Normalize Values of Alternatives versus Quantitative Criteria the value of an alternative Ai , i 1,2,..., m, versus a benefit quantitative criterion j, j g 1,..., h, and cost quantitative criterion j , j h 1,..., n . y ij (oij , pij , qij ) is rij denotes the normalized value of y ij rij ( oij pij qij , * , * ) , qij* max qij , j B * qij qij qij (4.3) oij oij oij rij ( * , * , * ) , o* min o , j C ij ij qij pij oij For calculation convenience, assume r ij (aij , bij , cij ) , j g 1,..., n . 39 The membership function of the final fuzzy evaluation value, Ti , i 1,..., n of each alternative can be developed as follows: g Ti w j xij j 1 h j g 1 w j xij n j h1 wj xij , (4.4) The membership functions are developed as: wj [(b j a j ) a j ,(b j c j ) c j ] , (4.5) xij [(eij dij ) dij ,(eij fij ) fij ] . (4.6) 40 From Eqs. (4.5) and (4.6), we can develop Eqs. (4.7) and (4.8) as follows: wj xij [(b j a j )(eij dij ) 2 (dij (b j a j ) a j (eij dij )) a j dij , (b j c j )(eij fij ) 2 ( fij (b j c j ) c j (eij fij )) c j fij ] . (4.7) n w j xij j 1 n n n j 1 j 1 n n j 1 j 1 [ (b j a j )(eij dij ) (dij (b j a j ) a j (eij dij )) a j dij , 2 j 1 n (b j c j )(eij fij ) ( fij (b j c j ) c j (eij fij ))) c j fij ] . j1 2 (4.8) 41 When applying Eq. (4.8) to Eq.(4.4), three equations are developed: g wj j 1 g g g j 1 j 1 xij [ (b j a j )(eij dij ) (dij (b j a j ) a j (eij dij )) a j dij , 2 j 1 g (b j c j )(eij fij ) g g j 1 j 1 ( fij (b j c j ) c j ( eij fij )) c j fij ] . 2 j1 (4.9) h j g 1 wj xij h [ j g 1 h j h1 wj xij [ (b j c j )(eij fij ) 2 j g 1 n (b j a j )(eij dij ) n j h1 n 2 j g 1 h j g 1 (b j a j )(eij dij ) 2 (b j c j )(eij fij ) jh1 h 2 (dij (b j a j ) a j (eij dij )) ( fij (b j c j ) c j (eij fij )) n j h1 n j h1 j g 1 a j dij , h j g 1 (dij (b j a j ) a j (eij dij )) ( fij (b j c j ) c j (eij fij )) h c j fij ] . n j h1 (4.10) a j dij , n j h1 c j fij ] . (4.11) 42 g Oi1 a j dij To simplify equation, assume: j 1 g Ai1 (b j a j )(eij dij ) j 1 Ai 2 Ai 3 j g 1 j h 1 (b j a j )(eij dij ) (b j a j )(eij dij ) g Bi1 dij (b j a j ) a j (eij dij ) j 1 Bi 2 Bi 3 h j g 1 n j h 1 Ci1 (b j c j )(eij fij ) j 1 h n g Ci 2 Ci 3 Oi 3 h j g 1 n j h 1 (b j c j )(eij fij ) (b j c j )(eij fij ) dij (b j a j ) a j (eij dij ) Di 3 j 1 h j g 1 n j h 1 fij (b j c j ) c j (eij fij ) j g 1 a j dij n j h 1 a j dij g Pi1 b j eij j 1 Pi 2 g Di1 fij (b j c j ) c j (eij fij ) dij (b j a j ) a j (eij dij ) D i2 Oi 2 h Pi 3 h j g 1 b j eij n j h 1 b j eij g Qi1 c j fij j 1 fij (b j c j ) c j (eij fij ) Qi 2 Qi 3 h j g 1 c j fij n j h 1 c j fij 43 By applying the above equations, Eqs. (4.9)-(4.11) can be arranged as Eqs. (4.12)-(4.14) as follows: g wj xij [ Ai1 2 Bi1 Oi1, Ci1 2 Di1 Qi1] . (4.12) j 1 h wj xij [ Ai 2 2 Bi 2 Oi 2 , Ci 2 2 Di 2 Qi 2 ] . (4.13) wj xij [ Ai 3 2 Bi 3 Oi 3 , Ci 3 2 Di 3 Qi 3 ] . (4.14) j g 1 n j h 1 44 Applying Eqs.(4.12)-(4.14) to Eq.(4.4) to produce Eq.(4.15): Ti [( Ai1 Ai 2 Ci 3 ) 2 ( Bi1 Bi 2 Di 3 ) (Oi1 Oi 2 Qi3 ) , (Ci1 Ci 2 Ai 3 ) ( Di1 Di 2 Bi 3 ) (Qi1 Qi 2 Oi3 )] . 2 Assume: (4.15) Ii1 Ai1 Ai 2 Ci 3 J i1 Bi1 Bi 2 Di 3 Ii 2 Ci1 Ci 2 Ai 3 J i 2 Di1 Di 2 Bi 3 Qi Oi1 Oi 2 Qi3 Yi Pi1 Pi 2 Pi 3 Zi Qi1 Qi 2 Oi 3 45 By applying the above equations, Eqs. (4.15) can be arranged as Eqs. (4.16) and (4.17) as follows: Ii1 2 J i1 Qi x 0 (4.16) Ii 2 2 Ji 2 Zi x 0 (4.17) The left and right membership function of Ti can be obtained as shown in Eq. (4.18) and (4.19) as follows: a 1 2 f J i1 J i12 4 Ii1 x Qi ,Q x Y ; x i i Ti 2 Ii1 f J i 2 J i 2 4 I i 2 x Zi ,Y x Z x i i Ti 2 Ii 2 L 2 a R (4.18) 1 2 (4.19) 46 Rank Fuzzy Numbers In this research, we applied total relative area to rank the final fuzzy numbers, and it is developed based on the concept of Chen’s (1985) maximizing set and minimizing set. 47 Using Chen’s maximizing set and minimizing set on the given sets of examples, we can see three rankings for the same alternatives. Table 4.1 Chen’s maximizing set and minimizing set ranking outcomes Example A1 A2 A3 A4 Ranking 1 (3, 5, 7) (4, 5, 51/8) (2, 3, 5) (8, 9, 10) A1=A2 2 (3, 5, 7) (4, 5, 51/8) (2, 3, 5) (6, 7, 8) A1<A2 3 (3, 5, 7) (4, 5, 51/8) (2, 3, 5) (10, 11, 12) A1>A2 To resolve the inconsistency problem above, modification was made to Chen’s ranking method. 48 Definition 1 The maximizing set M is a fuzzy subset with fM as f xRi1 xmin k ) , xmin xRi1 xmax , ( x xmax xmin M 0, otherwise . (4.20) The minimizing set N is a fuzzy subset with fN as f xLi1 xmax k ) , xmin xLi1 xmax , ( x xmin xmax N 0, otherwise , Where xmin infx S , xmax sup S , S i1 Si x is set to 1. n (4.22) , Si {x f Ai ( x) 0} , usually k 49 S ( xLi1 ) Yi xLi1 Yi f V x dx xLi1 S ( xRi 2 ) xRi 2 L ij S ( xLi 2 ) xLi 2 f Yi x dx xLi 2 Yi Vij R f V x dx Qi L Figure 4.1. Total Relative Area Zi S ( xRi1 ) ij ST ( Ai ) xL xmax i2 xmin xmax f V x dx xRi1 1 ( S ( xRi1 ) S ( xLi1 ) S ( xRi 2 ) S ( xLi 2 )) , i 1 ~ n . 4 R ij 50 By applying the new method, we can see that the rankings have changed and are now consistent. Table 4.4 Total relative area ranking outcomes Example A1 A2 A3 A4 Ranking 1 (3, 5, 7) (4, 5, 51/8) (2, 3, 5) (8, 9, 10) A1>A2 2 (3, 5, 7) (4, 5, 51/8) (2, 3, 5) (6, 7, 8) A1>A2 3 (3, 5, 7) (4, 5, 51/8) (2, 3, 5) (10, 11, 12) A1>A2 51 Compare the Chen’s (2010) maximizing area and minimizing area and the total relative area method by changing values of A2. Using Chen’s maximizing area and minimizing area on the given sets of examples, we can see three rankings for the same alternatives as shown in Table 4.6. Table 4.6 Maximizing area and minimizing area ranking outcomes Example A1 A2 A3 A4 Ranking 1 (3, 5, 7) (41/20, 5, 55/8) (2, 3, 5) (8, 9, 10) A1=A2 2 (3, 5, 7) (41/20, 5, 55/8) (2, 3, 5) (6, 7, 8) A1>A2 3 (3, 5, 7) (41/20, 5, 55/8) (2, 3, 5) (10, 11, 12) A1<A2 The Chen’s (2010) method have the inconsistency problem. 52 By applying the total relative area method, we can see that the rankings have changed and are now consistent as shown in Table 4.8. Table 4.8 Total relative area ranking outcomes Example A1 A2 A3 A4 Ranking 1 (3, 5, 7) (41/20, 5, 55/8) (2, 3, 5) (8, 9, 10) A1>A2 2 (3, 5, 7) (41/20, 5, 55/8) (2, 3, 5) (6, 7, 8) A1>A2 3 (3, 5, 7) (41/20, 5, 55/8) (2, 3, 5) (10, 11, 12) A1>A2 53 Using Liou and Wang method on the given sets of examples, we can see three rankings for the same alternatives as shown in Table 4.10. Table 4.10 The Liou and Wang method ranking outcomes Example A1 A2 A3 A4 Ranking 1 (3, 5, 7) (41/20, 5, 55/8) (2, 3, 5) (8, 9, 10) A1>A2 2 (3, 5, 7) (41/20, 5, 55/8) (2, 3, 5) (6, 7, 8) A1>A2 3 (3, 5, 7) (41/20, 5, 55/8) (2, 3, 5) (10, 11, 12) A1>A2 According to Table 4.10, the rankings are consistent; therefore, the feasibility of this model can be demonstrated. 54 xRi1 is developed as follows: xRi1 xmin xmax xmin 1 = J i 2 J i22 +4I i 2 xRi1 Zi 2 2I i 2 xRi1 1 2 xmax xmin J i 2 xmax xmin 2I i 2 xmin xmax xmin J i 2 xmax xmin +4I i 2 xmin Zi = 2Ii 2 2 (4.29) 55 xLi1 is developed as follows: xLi1 xmax xmin xmax xLi1 = xmin xmax J i1 1 = J i1 + J i21 +4I i1 xLi1 Qi 2 2I i1 xmin xmax 2Ii1xmax + xmin xmax J i1 xmin xmax +4I i1 xmin Qi 2 1 2 2Ii1 (4.30) 56 xRi 2 is developed as follows: xRi 2 xmin xmax xmin 1 = J i1 + J i21 +4I i1 xRi 2 Qi 2 2I i1 1 2 xmax xmin J i1 xmax xmin 2Ii1xmin + xmax xmin J i1 xmax xmin +4I i1 xmin Qi xRi 2 = 2Ii1 2 (4.33) 57 xLi 2 is developed as follows: xLi 2 xmax xmin xmax xLi 2 xmin xmax J i 2 1 = J i 2 J i22 +4I i 2 xLi 2 Zi 2 2I i 2 xmin xmax 2Ii 2 xmax xmin xmax J i 2 xmin 2Ii 2 xmax +4I i 2 xmin Zi 2 1 2 (4.34) 58 By equations(4.29),the S xRi1 area, denoted as: S xRi1 J i 2 Zi xRi1 2 Ii 2 J i32 12 Ii22 3 2 1 2 J i 2 4 Ii 2 xRi1 Zi 12 Ii 2 (4.31) xRi1 =equations(4.29) 59 By equations(4.30),the S xLi1 Yi xLi1 J i1 Yi xLi1 2 Ii1 area, denoted as: S xLi1 3 3 1 2 1 2 2 J 4 I Y Q J 4Ii1 xLi1 Qi 2 i1 i i 2 i1 2 i1 12 Ii1 12 Ii1 (4.32) xLi1 =equations(4.30) 60 By equations(4.33),the S xRi 2 area, denoted as: S xRi 2 J i1 xRi 2 Qi 2 Ii1 3 2 J i31 1 2 J 4 Ii1 xRi 2 Qi 2 i1 12 I 2 12 Ii1 i1 (4.35) xRi 2 =equations(4.33) 61 By equations(4.34),the S xLi 2 area, denoted as: S xLi 2 J i 2 xLi 2 Yi 2Ii 2 1 3 2 3 2 1 J i22 +4I i 2 xL Zi J i22 +4I i 2 Yi Z i i2 12 Ii22 12 Ii22 xL2 xL xmax Yi Yi xmax i2 i2 xmin xmax (4.36) xLi 2 =equations(4.34) 62 By equations (4.31),(4.32),(4.35)and(4.36), the total area, denoted as: ST ( Ai ) S ( xRi1 ) S ( xLi1 ) S ( xRi 2 ) S ( xLi 2 ) 4 ,i 1 ~ n . (4.37) 63 NUMERICAL EXAMPLE Averaged Weights of Criteria Ratings of Alternatives versus Qualitative Criteria Normalization of Quantitative Criteria Development of Membership Function Defuzzification 64 Assume that a user is looking for a suitable new smart phone. Suppose three professional persons, D1, D2 and D3, who have the knowledge background in smart phone. The five candidate smart phone are A1 (Nokia), A2 (Apple), A3 (HTC), A4 (Sony Ericsson) and A5 (RIM) . 65 Criteria Qualitative Brand awareness of smart phone Quantitative Benefit Cost Screen size Price of applications Camera resolution Price Number of applications Weight Brand awareness of OS Designs Security Operability Entertainment Execution efficiency Qualitative 66 Figure 5.1. Evaluation Criteria Table 5.1 Linguistic Values and Fuzzy Numbers for Ratings Linguistic values Triangular fuzzy numbers Very Unsatisfactory (0.0 ,0.1,0.3), Unsatisfactory (0.1,0.3,0.5) Ordinary (0.3,0.5,0.7) Satisfactory (0.5,0.7,0.9) Very Satisfactory (0.7,0.9,1.0) Table 5.2 Linguistic Values and Fuzzy Numbers for Weights Linguistic values Triangular fuzzy numbers Unimportant (0.0 ,0.1,0.3), Ordinary Important (0.1,0.3,0.5) Important (0.3,0.5,0.7) Very Important (0.5,0.7,0.9) Absolutely Important (0.7,0.9,1.0) 67 The Candidate Smart Phone Table 5.3 The Candidate Smart Phones Manufacrer Sony Ericsson RIM XPERIA BlackBerry Arc Bold 9780 Windows Phone 7 Android 2.3 BlackBerry OS 6.0 3.5 inches 3.7 inches 4.2 inches 2.44 inches 8 million pixels 5 million pixels 8 million pixels 8.1 million pixels 5 million pixels 640*360 640*960 480*800 480*854 480*360 pixels pixels pixels pixels pixels NT. 20400 NT.22200 NT.11200 NT.18600 NT.15000 Nokia Apple HTC E7 iPhone 4 7 Mozart Symbian^3 iOS 4 4.0 inches camera resolution screen Smart Phone Type OS Screen size resolution Price 68 Weight 176 g 137 g 130 g 117 g 122 g Averaged Weights of Criteria Assume that the importance weights given by decision makers to each criterion are shown in Table 5.4. Table 5.4 Linguistic Weights of Criteria DM D1 D2 D3 C2 AI OI VI VI I I C3 I VI AI C4 AI AI AI UI VI I AI I OI UI UI VI AI VI I C12 UI OI OI I I I I VI OI VI UI AI C13 I VI OI Criteria C1 C5 C6 C7 C8 C9 C10 C11 69 According to Table 5.2, the corresponding triangular fuzzy numbers are shown in Table 5.5. Table 5.5 Fuzzy Weights of Criteria DM D1 D2 C1 (0.7,0.9,1.0) (0.5,0.7,0.9) C2 (0.1,0.3,0.5) (0.5,0.7,0.9) (0.3,0.5,0.7) C3 (0.3,0.5,0.7) (0.5,0.7,0.9) (0.7,0.9,1.0) C4 (0.7,0.9,1.0) (0.3,0.5,0.7) (0.0,0.1,0.3) C5 (0.7,0.9,1.0) (0.7,0.9,1.0) (0.5,0.7,0.9) C6 (0.7,0.9,1.0) (0.3,0.5,0.7) (0.7,0.9,1.0) C7 (0.0,0.1,0.3) (0.1,0.3,0.5) (0.5,0.7,0.9) C8 (0.5,0.7,0.9) (0.0,0.1,0.3) (0.3,0.5,0.7) C9 (0.0,0.1,0.3) (0.3,0.5,0.7) (0.1,0.3,0.5) C10 (0.1,0.3,0.5) (0.3,0.5,0.7) (0.5,0.7,0.9) C11 (0.1,0.3,0.5) (0.3,0.5,0.7) (0.0,0.1,0.3) C12 (0.3,0.5,0.7) (0.5,0.7,0.9) (0.7,0.9,1.0) C13 (0.3,0.5,0.7) (0.5,0.7,0.9) (0.1,0.3,0.5) Criteria D3 (0.3,0.5,0.7) 70 By Eq. (4.1), the averaged weight of each criterion can be obtained as shown in Table 5.6. Table 5.6 Averaged Weights of Criteria Criteria Averaged Weights C1 0.5 0.7 0.86 C2 0.3 0.5 0.7 C3 0.5 0.7 0.86 C4 0.33 0.5 0.66 C5 0.63 0.83 0.96 C6 0.56 0.76 0.9 C7 0.2 0.36 0.56 C8 0.26 0.43 0.63 C9 0.13 0.3 0.5 C10 0.3 0.5 0.7 C11 0.13 0.3 0.5 C12 0.5 0.7 0.86 C13 0.3 0.5 0.7 71 Ratings of Alternatives versus Qualitative Criteria According to the step in Section 4.2, we obtain the ratings of alternatives versus qualitative criteria as shown in Table 5.7. Table 5.7 Ratings of Smart Phone Candidates versus Qualitative Criteria A1 Criteria C1 Candidates A1 O S O VS A3 O S VS O A4 U S O D1 D3 VS S C4 A2 S VS S O O VS A3 VS O VS A1 O O VS O A2 VS S VS A3 VS VS VS A4 O S O A5 O O O A1 VS S S S S O VS S O O A5 A1 C3 VS A5 A4 C2 S A2 Decision Makers D2 Smart Phone O S S O O O O C5 O A2 VS O VS A3 S O S A4 VS S O A5 VS S VS A1 U O U A5 S S VS S O A2 VS VS O A1 U O VU A3 U VS O A2 O O O A4 S O U A3 U O O A5 S S S A4 U VS O A5 U O O A2 C6 A3 A4 C7 72 By Eq. (4.2), the averaged fuzzy evaluation values of candidates versus qualitative criteria can be obtained as shown in Table 5.8. Table 5.8 Averages Fuzzy Evaluation Values of Candidates versus Qualitative Criteria Criteria C2 C1 C3 C4 C5 C6 C7 Alternative A1 A2 A3 A4 A5 0.5 0.36 0.16 0.5 0.43 0.5 0.13 0.7 0.56 0.36 0.7 0.63 0.7 0.3 0.86 0.76 0.56 0.86 0.8 0.86 0.5 0.56 0.56 0.56 0.36 0.63 0.56 0.3 0.76 0.76 0.76 0.56 0.83 0.76 0.5 0.93 0.9 0.9 0.76 0.96 0.93 0.7 0.56 0.43 0.36 0.43 0.7 0.56 0.3 0.76 0.63 0.56 0.63 0.9 0.76 0.5 0.9 0.83 0.73 0.83 1 0.93 0.7 0.3 0.5 0.3 0.56 0.36 0.43 0.36 0.5 0.7 0.5 0.76 0.56 0.63 0.56 0.7 0.86 0.7 0.93 0.76 0.83 0.73 0.36 0.63 0.5 0.43 0.3 0.36 0.23 0.56 0.83 0.7 0.63 0.5 0.56 0.43 0.76 0.96 0.9 0.8 0.7 0.76 0.63 73 Normalization of Quantitative Criteria rij ( The suitability values of candidates versus quantitative criteria can be shown in Table 5.9. oij pij qij , * , * ) , qij* max qij , j B * qij qij qij Criteria A1 A2 Candidates A3 A4 A5 C8 (3.9, 4.0, 4.1) (3.2, 3.5, 3.7) (3.5, 3.7, 4.0) (4.1, 4.2, 4.5) (2.25, 2.44, 2.64) C9 (730, 800, 870) (445, 500, 555) (720, 800, 880) (729, 810, 891) (450, 500, 550) C10 (6.0,6.5,6.7) (17.0,17.5,18.0) (5.5,6.0,6.5) (4.4,4.8,5.0) (4,4.2,4.5) C11 (6.4,6.6,6.8) (3.2,3.4,3.6) (3.0,3.2,3.4) (2.8,3.0,3.2) (7.8,8.0,8.2) C12 (19000,20400,20900) (21800,22200,23000) (10200,11200,12000) (18000,18600,19000) (14500,15000,15500) C13 (173, 176, 179) (134, 137, 140) (125, 130, 135) (115, 117, 119) (120, 122, 125) oij oij oij rij ( * , * , * ) , o* min o , j C ij ij qij pij oij 74 By Eq. (4.3), the normalized values of candidates versus quantitative criteria can be obtained as shown in Table 5.10. Table 5.10 Normalization of Quantitative Criteria Criteria C8 C9 C10 C11 C12 C13 Smart Phone Candidates Normalization of Quantitative Criteria A1 0.8666 0.8888 0.9111 A2 0.7111 0.7777 0.8222 A3 0.7777 0.8222 0.8888 A4 0.9111 0.9333 1 A5 0.5 0.5422 0.5866 A1 0.8193 0.8978 0.9764 A2 0.4994 0.5611 0.6228 A3 0.8080 0.8978 0.9876 A4 0.8181 0.9090 1 A5 0.5050 0.5611 0.6172 0.3888 A1 0.3333 0.3611 A2 0.9444 0.9722 1 A3 0.3055 0.3333 0.3611 A4 0.2444 0.2666 0.2777 A5 0.2222 0.2333 0.25 A1 0.4375 0.4242 0.4117 A2 0.875 0.8235 0.7777 0.8235 A3 0.9333 0.875 A4 1 0.9333 0.875 A5 0.3589 0.35 0.3414 A1 0.5368 0.5 0.4880 A2 0.4678 0.4594 0.4434 A3 1 0.9107 0.85 A4 0.5666 0.5483 0.5368 A5 0.7034 0.68 0.6580 A1 0.6647 0.6534 0.6424 A2 0.8582 0.8394 0.8214 A3 0.92 0.8846 0.8518 A4 1 0.9829 0.9663 A5 0.9583 0.9426 0.92 75 Development of Membership Function By Eqs. (4.4)-(4.19), the membership function, Ti, of the final fuzzy evaluation value of each smart phone candidate can be developed through α-cut and interval arithmetic operations. 76 g Oi1 a j dij g j 1 Ai1 (b j a j )(eij dij ) j 1 h Ai 2 j g 1 n Ai 3 j h 1 (b j a j )(eij dij ) Bi1 dij (b j a j ) a j (eij dij ) j 1 h j g 1 n Bi 3 j h 1 dij (b j a j ) a j (eij dij ) dij (b j a j ) a j (eij dij ) g Ci1 (b j c j )(eij fij ) j 1 Ci 2 Ci 3 h j g 1 n j h 1 (b j c j )(eij fij ) (b j c j )(eij fij ) g Di1 fij (b j c j ) c j (eij fij ) j 1 Di 2 Di 3 h j g 1 n j h 1 fij (b j c j ) c j (eij fij ) Oi 2 , Oi1, Oi 2 , Oi 3, Pi1, Pi 2 , Pi 3, Qi1, Qi 2 , Qi 3 (b j a j )(eij dij ) g Bi 2 Table 5.11 Values for Ai1, Ai 2 , Ai 3, Bi1, Bi 2 , Bi 3, Ci1, Ci 2 , Ci 3, Di1, Di 2 , Di 3 A1 A2 A3 A4 A5 Ai1 Ai 2 Ai 3 0.261 0.266 0.266 0.266 0.266 0.022 0.026 0.027 0.023 0.018 -0.01 -0.014 -0.034 -0.018 -0.009 Bi1 Bi 2 Bi 3 1.098 1.297 1.255 1.142 1.151 0.372 0.294 0.424 0.395 0.357 0.488 0.361 0.459 0.234 0.377 Ci1 Ci 2 Ci 3 Di1 Di 2 Di 3 0.213 0.190 0.198 0.214 0.214 0.113 0.100 0.026 -0.015 0.036 -0.028 0.033 -0.017 0.023 -0.010 -1.852 -1.885 -1.868 -1.964 -1.964 -0.749 -0.655 -0.509 -0.392 -0.484 -0.475 -0.473 -0.521 -0.313 -0.367 Oi1 Oi 2 Oi 3 1.194 1.641 1.565 1.190 1.212 0.440 0.454 0.539 0.545 0.406 0.767 0.425 0.641 0.267 0.593 Pi1 Pi 2 Pi 3 Qi1 Qi 2 Qi 3 2.554 0.788 3.205 0.930 3.087 0.732 2.598 0.762 2.630 0.479 0.767 0.956 1.306 1.121 0.985 4.193 1.191 4.901 1.361 4.757 1.110 4.348 1.157 4.380 0.724 1.095 1.396 1.877 1.634 1.389 Oi 3 h j g 1 a j dij n j h 1 a j dij g Pi1 b j eij j 1 Pi 2 Pi 3 h j g 1 b j eij n j h 1 b j eij g Qi1 c j fij j 1 Qi 2 Qi 3 h j g 1 c j fij n j h 1 c j fij fij (b j c j ) c j (eij fij ) 77 Table 5.12 Values for Ai1 Ai 2 Ci 3, Bi1 Bi 2 Di 3, Oi1 Oi 2 Qi 3, , Ci1 Ci 2 Ai 3, Di1 Di 2 Bi 3, Pi1 Pi 2 Pi 3, Qi1 Qi 2 Oi 3 Ai1 Ai 2 Ci 3 Bi1 Bi 2 Di 3 Oi1 Oi 2 Qi 3 Ci1 Ci 2 Ai 3 Di1 Di 2 Bi 3 Pi1 Pi 2 Pi 3 Qi1 Qi 2 Oi 3 A1 A2 A3 A4 A5 0.19 2.23 0.31 2.12 0.32 2.10 0.31 2.04 0.28 1.60 0.91 0.30 1.01 0.23 0.34 0.27 0.25 0.26 0.02 0.25 -2.86 3.10 -2.74 3.41 -2.79 2.76 -2.84 2.51 -2.52 1.89 5.48 5.89 5.29 5.07 4.13 Ti [( Ai1 Ai 2 Ci 3 ) 2 ( Bi1 Bi 2 Di 3 ) (Oi1 Oi 2 Qi3 ) , (Ci1 Ci 2 Ai 3 ) ( Di1 Di 2 Bi 3 ) (Qi1 Qi 2 Oi3 )] . 2 (4.15) 78 L R f By Eq. (4.18)and(4.19), the left, Ti ( x), and right, fT ( x) , membership functions of the final fuzzy evaluation value Ti can be obtained as shown in Table 5.13. i 79 80 Final Ranking Values Using the suggested total relative area method, we can obtain S xR , S xL , S xR and S xL using Eq.(4.31) , (4.32), (4.35) and (4.36) respectively, and the total area value can be obtained by using Eq.(4.37), as shown in Table 5.15. i1 i1 i2 i2 81 Table 5.15 S xRi1 , S xRi 2 , S xLi1 , S xLi 2 and ST Ai of Each Alternative S(x Ri1 ) A1 A2 A3 A4 A5 34.235 48.762 38.451 41.085 30.423 S(x Ri2 ) A1 A2 A3 A4 A5 0.070 0.082 0.008 0.004 0.000 S(x Li1 ) A1 A2 A3 A4 A5 0.184 0.210 0.135 0.116 0.058 S(x Li2 ) A1 A2 A3 A4 A5 6.160 8.035 5.263 4.343 2.121 ST(Ai) A1 A2 A3 A4 A5 10.162 14.272 10.964 11.387 8.150 RANK A1 A2 A3 A4 A5 4 1 3 2 5 82 According to Table 5.15 the ranking order of the five candidate smart phones is A2 > A4 > A3 > A1 > A5 . A2 has the largest area 14.272 ; therefore, A2 is the most suitable smart phone for decision makers under the evaluation procedure of the proposed model. 83 CONCLUSIONS Conclusions Suggestions for Future Research 84 Conclusions 1. The needed criteria for evaluating smart phones are carefully analyzed and selected through thoroughly comprehensive investigation on relevant literature survey. 2. This research uses a fuzzy MCDM to develop an evaluation and selection model for smart phones. 3. Ranking formulae are clearly developed for better executing the decision making. 4. A numerical example is used to demonstrate the feasibility of the proposed fuzzy MCDM model. 85 Suggestions for Future Research 1. In this research, only one level in the criteria hierarchical structure is considered. In future research, a multiple levels may be further investigated for better depicting the relationship in the criteria hierarchical structure for smart phone. 2. In the numerical example, ratings of alternatives and importance weights of criteria are subjectively assigned by decision makers. Some objective methods, such as survey, can be used to strengthen the effectiveness of the proposed model. 86 3. In fuzzy number ranking, this research suggests the total relative area based on Chen (1985) maximizing set and minimizing set for defuzzification. In future research, some other ranking approaches may be used for the proposed fuzzy MCDM model. However, the ranking results may be different. 4. Fuzzy numbers other than triangular can also be used for the proposed model, a comparison may be needed. 5. The linguistic values and their corresponding fuzzy numbers used in this research can be adjusted to fit different applications. A model may be needed to objectively produce these values. 87 88