324 JOURNAL OF NETWORKS, VOL. 4, NO. 5, JULY 29 Fuzzy Evaluation on Networ Security Based on the New Algorith of Mebership Degree Transforation M(,2,3) Hua Jiang School of Econoics and Manageent, Hebei University of Engineering, Handan, China hdianghua@26.co Junhu Ruan School of Econoics and Manageent, Hebei University of Engineering, Handan, China ruanunhu@63.co Abstract Networ security has close relationships with the physical environent of inforation carriers, inforation transission, inforation storage, inforation anageent and so on, and these relations have uch abiguity. Therefore, it is reasonable and scientific to apply fuzzy coprehensive evaluation ethod for networ security fuzzy evaluation. The core of networ security fuzzy evaluation is ebership degree transforation calculation. But the transforation ethods should be discussed, because redundant data in index ebership degree is also used to copute obect ebership degree, which is not useful for obect classification. The new algorith is: using data ining technology based on entropy to ine nowledge inforation about obect classification hidden in every index, affir the relationship of obect classification and index ebership, eliinate the redundant data in index ebership for obect classification by defining distinguishable weight and extract valid values to copute obect ebership. The new algorith of ebership degree transforation includes three calculation steps which can be suarized as effective, coparison and coposition, which is denoted as M(,2,3). The paper applied the new algorith in networ security fuzzy evaluation. Index Ters networ security, ebership degree transforation, fuzzy evaluation, M(,2,3) odel I. INTRODUCTION With the developent and universal application of coputer networs and counication technologies, our lives have been undergoing enorous changes. Internet and odern counications have brought us great convenience and speed. Our lives and wors are increasingly dependent on the, which provide us a variety of inforation. But at the sae tie, a new proble is also haunting us networ security. Networ security probles can be found in any areas such as coputer systes subected to virus infection and daage, coputer hacing activities, online political Hua Jiang, 977--9, Handan, China, hdianghua@26.co subversion activities, and so on. In view of this status quo, networ construction and aintenance personnel should tae preventive easures to solve networ security probles and iniize various networ security threats, so it is essential to evaluate networ syste security in advance. There are any factors that effect networ security such as hardware equipent, software systes, an-ade destruction, anageent syste, external environent and so on, which ultiately deterine the safety of networs together. And the relations aong these factors have uch abiguity, so it is reasonable and scientific to apply fuzzy coprehensive evaluation ethod for networ security fuzzy evaluation. Ref. [], according to the principles of scientificity, coprehensiveness, feasibility and coparability, set up a networ security evaluation index syste and used expert scoring ethod to deterine the ebership vectors of the base indexes on five evaluation grades, which fored networ security fuzzy evaluation atrix, as Table shows. The core of networ security fuzzy evaluation is ebership degree transforation calculation. But the existing transforation ethods including the ethod in Ref. [] should be discussed, because redundant data in index ebership degree is also used to copute obect ebership degree, which is not useful for obect classification. The new algorith is: using data ining technology based on entropy to ine nowledge inforation about obect classification hidden in every index, affir the relationship of obect classification and index ebership, eliinate the redundant data in index ebership for obect classification by defining distinguishable weight and extract valid values to copute obect ebership. The paper will apply the new algorith in networ security fuzzy evaluation. II. EXISTING MEMBERSHIP DEGREE TRANSFORMATION METHODS A. Fuzzy Logic The theoretical basis of fuzzy logic is fuzzy set theory, 29 ACADEMY PUBLISHER
JOURNAL OF NETWORKS, VOL. 4, NO. 5, JULY 29 325 which was first proposed by Lotfi Zadeh in 965 (Stefi 995) [2]. Fuzzy set theory was introduced to solve probles that are ipossible for classical set theory and two-valued logic. In the real world, people possess extensive abilities to deal with fuzzy nowledge, which ay be vague, iprecise, uncertain, abiguous, inexact or probabilistic in nature (Orchard 995). People are also able to reason and solve probles using this fuzzy nowledge. While it is difficult for traditional logic to represent these fuzzy concepts and siulate the fuzzy reasoning process, fuzzy logic overcoes this liitation by extending classical set theory and logic [3]. In traditional set theory, an eleent x in the universal set U either belongs to a set S or does not. Fuzzy set theory, on the other hand, allows an eleent x in universal set U to partially belong to a fuzzy set FS. A fuzzy set can be described by a characteristic function or ebership function. The ebership value of an eleent in that fuzzy set can vary fro. to.. A ebership value of. indicates that the eleent x has no ebership in the fuzzy set FS. On the other hand, a ebership value of. indicates that x has coplete ebership in FS. Using this idea, any fuzzy concepts can be represented readily. For exaple, suppose the universe of discourse of huan age is between and ; then the fuzzy concepts Young and Old can be expressed graphically as shown in Fig.. One point noticeable in fuzzy logic is that an eleent x can belong to a given fuzzy set S and its copleent S at the sae tie. This characteristic does not hold in traditional two-valued logic [4]. Fro Fig., it can be seen that when one is less than 2, this person is young (ebership is.). When this person gets older, his/her degree of being young decreases. Last, when he/she reaches the age 55 or so, this person is no longer young (ebership is ). The curve representing the fuzzy set Old can be explained siilarly. Consequently, a fuzzy variable can tae these fuzzy sets as its values. For exaple, if John s age is a fuzzy variable, it can tae on the values Young or Old. In practice two ost frequently adopted ethods to represent a ebership function are enueration representation and function representation. In certain situations, it is convenient to use a set of strict atheatical functions lie those illustrated in Fig. 2 to represent the ebership functions of a fuzzy set. Another ore coonly used approach is to siplify the Figure 2. Linear Function Representation of Fuzzy Set Young strict atheatical function by using a piecewise linear function (see Fig. 2 for an exaple) and to further represent the function by enueration. For exaple piecewise linear function for Young shown in Fig. 2 also could be expressed as: Young=(/., 2/., 3/.6, 4/.2, 5/., 6/., /.) B. Mebership Functions The ebership function of a fuzzy set is a generalization of the indicator function in classical sets. In fuzzy logic, it represents the degree of truth as an extension of valuation. Degrees of truth are often confused with probabilities, although they are conceptually distinct, because fuzzy truth represents ebership in vaguely defined sets, not lielihood of soe event or condition. Mebership functions were introduced by Zadeh in the first paper on fuzzy sets (965). For any set X, a ebership function on X is any function fro X to the real unit interval [, ]. Mebership functions on X represent fuzzy subsets of X. The ebership function which represents a fuzzy set A ~ is usually denoted by A. For an eleent x of X value A (x) is called the ebership degree of x in the fuzzy set A ~. The ebership degree A (x) quantifies the grade of ebership of the eleent x to the fuzzy set A ~. The value eans that x is not a eber of the fuzzy set; the value eans that x is fully a eber of the fuzzy set. The values between and characterize fuzzy ebers, which belong to the fuzzy set only partially [5]. Mebership function of a fuzzy set is shown as Fig. 3. Soeties [2], a ore general definition is used, where ebership functions tae values in an arbitrary fixed algebra or structure L; usually it is required that L be at least a poset or lattice. The usual ebership Figure. Graphical Representation of the Fuzzy Sets Young and Old Figure 3. Mebership function of a fuzzy set 29 ACADEMY PUBLISHER
326 JOURNAL OF NETWORKS, VOL. 4, NO. 5, JULY 29 functions with values in [,] are then called [,] valued ebership functions. C. Existing Mebership Degree Transforation Methods For a hierarchical structure, if obtaining ebership degree of i index belonging to C class, it can obtain ebership degree fro interediate level to top general goal Z belonging to C class. And every ebership degree transforation in every level can be suarized in the following ebership transforation odel: Suppose that there are indexes which affect obect Q, where the iportance weights λ of ( = ~ ) index about obect Q is given and satisfies: λ, = Q = λ () Every index is classified into p classes. C K represents the K th class and C K is prior to C K+. If the ebership K (Q ) of th index belonging to C K is given, where K = ~ P and = ~, and K (Q ) satisfies:, = (2) K P K = What is the ebership K (Q) of obect Q belonging to C K? Obviously, whether the above transforation ethod is correct or not deterines that the evaluation result is credible or not. For the above ebership transforationre are four transforation ethods in fuzzy coprehensive evaluation: M ( Λ, V ), M (, V ), M ( Λ, ) and M (, + ) [6]. However through a long-tie research on the application, only M (, + ) is accepted by ost researchers, which regards obect ebership as weighted su : ~ = = λ,( = p) (3) And the M (, + ) ethod as the ainstrea ebership transforation algorith is widely used [7-2]. And above ethod is basic ethod realizing ebership transforation fro universeu fuzzy set to universe V fuzzy set in fuzzy logical syste. But M (, + ) ethod is in dispute in acadeic circles especially in application field. For exaple, [3, 4] pointed out that the weighted su ethod was too siple and did not use inforation sufficiently. In [3], the authors proposed a subective and obective coprehensive ethod based on evidence deduction and rough sets theory to realize ebership transforation. In [4], in the iproved fuzzy coprehensive evaluation, a new coprehensive weight is defined to copute weighted su instead of index iportance weight. K However, including these entioned ethods, any existing ebership transforation ethods are not designed for obect classification, thus they can t indicate which parts in index ebership are useful for obect classification and which parts are useless. The redundancy of ebership degree transforation shows that: the correct ethod realizing ebership degree transforation is not found, which need further study. In order to delete the redundant data in existing ebership transforation ethods, we use data ining technology based on entropy [5-9] to ine nowledge inforation about obect classification hidden in every index, affir the relationship of obect classification and index ebership, eliinate the redundant data in index ebership for obect classification by defining distinguishable weight and extract valid values to copute obect ebership. III. THE NEW ALGORITHM OF MEMBERSHIP DEGREE TRANSFORMATION M(,2,3) Fro the viewpoint of classification, what are concerned ost are these following questions: Dose every index ebership play a role in the classification of obect Q? Are there redundant data in index ebership for the classification of obect Q? These questions are very iportant. Because their answers decide which index ebership and which value are qualified to copute ebership of obect Q. To find the answers, we analyze as follows. A. The Distinguishable Weight Assue that = 2 = L = p n th index ebership iplies that the probability of classifying obect Q into every grade is equal. Obviously, this inforation is of no use to the classification of obect Q. Deleting th index will not affect classification. Let α (Q) represent the noralized and quantized value describing th index contributes to classificationn in this case α =. 2 If there exists an integer K satisfying = and other eberships are zeron th index ebership iplies that Q can be only classified into C. In this case, th index contributes ost to classification and α (Q) should obtain its axiu value. 3 Siilarly, if (Q) is ore concentrated for K, th index contributes ore to classification, i.e., α (Q) is larger. Conversely, if (Q) is ore scattered for K, th index contributes less to classification, i.e., α (Q) is saller. The above ()~(3) show that α (Q), reflecting the value that th index contributes to classification, is decided by the extent (Q) is concentrated or scattered for K. And it can be described quantitatively 29 ACADEMY PUBLISHER
JOURNAL OF NETWORKS, VOL. 4, NO. 5, JULY 29 327 by the entropy H (Q). Therefore, α (Q) is a function of H (Q) : p H = log = (4) v = H (5) log p t t= α = ν ν ( = ~ ) (6) Definition : If (Q) ( = ~ p, = ~ ) is the ebership of th index belonging to C and satisfies Eq. (); by (4) (5) (6), α (Q) is called distinguishable weight of th index corresponding to Q. Obviously, α (Q) satisfies B. The Effective Value α, α = (7) = The significance of α (Q) lies in its distinguishing function, i.e., it is a easure that reveals the exactness of obect Q being classified by th index ebership and even the extent of the exactness. If α =, fro the properties of entropyn = 2 = L = p. This iplies th index ebership is redundant and useless for classification. Naturally the redundant index ebership can t be utilized to copute ebership of obect Q. Definition 2: If (Q) ( = ~ p, = ~ ) is the ebership of th index belonging to C and satisfies Eq. (), and α (Q) is the distinguishable weight of th index corresponding to Q n α ( = ~ p) (8) is called effective distinguishable value of K th class ebership of th index, or K th class effective value for short. If α =, it indicates that th index ebership is redundant and useless for the classification of obect Q, so it can not be utilized to copute ebership of obect Q. Note that if α = n α =. So in fact coputing K th class ebership (Q) of obect Q isn t to find (Q) but to find α. This is a crucial fact. When index ebership is replaced by effective value to copute obect ebership, distinguishable weight is a filter. In the progress of ebership transforation, it can delete the redundant index eberships that are useless in classification and the redundant values in index ebership. C. The Coparable Value Undoubtedly, α is necessary for calculating (Q). However the proble is in general K th class effective values of different indexes aren t coparable and can t be added directly. Because, for deterining K th class ebership of obect Q, in ost cases these effective values are different in unit iportance. The reason is, generally, index ebership doesn t iply relative iportance of different indexes. So when using K th class effective value to copute K th class ebership, K th effective value ust be transfored into K th class coparable effective value. Definition 3: If α is K th class effective value of th index, and β (Q) is iportance weight of th index related to obect Q n β α ( = ~ p) (9) is called coparable effective value of K th class ebership of th index, or K th class coparable value for short. Clearly, K th class coparable values of different indexes are coparable between each other and can be added directly. Definition 4: If β α is K th class coparable value of th index of Q, where ( = ~ ), then = M = β α ( = ~ p) () is naed K th class coparable su of obect Q. Obviously bigger M (Q) is ore possibly that obect Q belongs to C K. Definition 5: If M (Q) is K th class coparable su of obect Q, and (Q) is the ebership of obect Q belonging to n Δ C K p = M M ( = ~ p) () t= Obviously, given by Eq.(), ebership degree (Q) satisfies: t p = (2) = Up to now, supposing that index ebership and index iportance weight are given, by Eq. (4), (5), (6), (8), (9), (), () transforation fro index ebership to obect ebership is realized. And this transforation needs no prior nowledge and doesn t cause wrong classification inforation. The above ebership transforation ethod can be suarized as effective, coparison and coposition, M,2,3 [2]. which is denoted as 29 ACADEMY PUBLISHER
328 JOURNAL OF NETWORKS, VOL. 4, NO. 5, JULY 29 TABLE I. THE INDEX DATA OF NETWORK SECURITY FUZZY EVALUATION The goal The criterion level The factor level Mebership vectors (C, C 2, C 3, C 4, C 5 ) Networ security fuzzy evaluation S Physical security A (.33) Logical security A 2 (.46) Manageent security A 3 (.2) Surrounding roo environent B (.7) (.2,.6,.2,.,.) Teperature, huidity and cleanliness control B 2 (.3) (.5,.5,.,.,.) Fire prevention easures B 3 (.9) (.6,.4,.,.,.) Lightning prevention easures B 4 (.3) (.3,.6,.,.,.) Anti-static easures B 5 (.) (.7,.3,.,.,.) Power supply B 6 (.3) (.9,.,.,.,.) Free-standing equipent on ground and des fixed easures B 7 (.) (.,.3,.6,.,.) Heart touching settings B 8 (.8) (.,.6,.2,.,.) Servers, bacbone equipent and lines bacup strategies B 9 (.5) (.,.5,.4,.,.) Bacup of iportant data B 2 (.9) (.6,.4,.,.,.) Syste security audit B 22 (.8) (.3,.5,.2,.,.) Syste operation log B 23 (.6) (.,.2,.7,.,.) Data recovery echanis B 24 (.6) (.,.,.2,.5,.2) Data encryption B 25 (.3) (.8,.2,.,.,.) Inforation syste access control echaniss B 26 (.) (.6,.3,.,.,.) Syste software security B 27 (.4) (.2,.6,.2,.,.) Anti-hacing easures B 28 (.7) (.,.3,.4,.3,.) Anti-virus easures B 29 (.7) (.4,.4,.2,.,.) Inforation anageent security sector leaders and organization B 3 (.8) (.9,.,.,.,.) Inforation security syste B 32 (.2) (.4,.5,.,.,.) Inforation security personnel training easures B 33 (.9) (.2,.4,.3,.,.) Equipent and data anageent syste copleteness B 34 (.2) (.6,.4,.,.,.) Registration file syste B 35 (.) (.9,.,.,.,.) Anti-theft and anti-loss easures B 36 (.) (.,.,.2,.6,.2) Password anageent security syste B 37 (.7) (.3,.3,.4,.,.) Accident eergency plan B 38 (.) (.6,.3,.,.,.) IV. FUZZY EVALUATION ON NETWORK SECURITY BASED ON M(,2,3) A. Networ Security Fuzzy Evaluation Matrix According to Ref. [], we can the networ security fuzzy evaluation atrix in soe organization, as Table I shows. In Table I values in bracets behind corresponding indexes are their iportance weights; the vectors behind the base indexes are their ebership vectors including five grades: C (very good), C 2 (good), C 3 (general), C 4 (bad), and C 5 (very bad). The data in the table are fro Ref. []. B. Fuzzy Evaluation Steps Based on M(,2,3) Model ) Calculating the ebership vector of physical security A : A includes nine base indexes B ~ B9 evaluation atrix is:.2.6.5.5.6.4.3.6 U ( A ) =.7.3.9...3..6..5.2..................6...2...4.. 29 ACADEMY PUBLISHER
JOURNAL OF NETWORKS, VOL. 4, NO. 5, JULY 29 329 According to the th row ( = ~ 9) A distinguishable weights of B are obtained and the distinguishable weight vector is: α A = (.89.96.237.73.265.9) of U.96.349.735 2 In Table I iportance weight vector of B ~ B 9 is given: β ( A ) = (.7.3.9.3..3..8.5 ) 3 Calculate the K th coparable value of =,2 L9 and obtain the coparable value atrix B N ( A ) of A :.2.8.68.37 N( A ) =.4.23.6.3.37.8.46.75.45.23.32.34.67.2.2.63..54..6 4 According to N ( A ), calculate the K th coparable su of A and obtain the coparable su vector: M ( A ) = (.525.438.54.6 ) 5 According to M ( A ), calculate the ebership vector ( A ) of A : ( A ) = (.463.387.356.43 ) 2) Calculating the ebership vector of logical security A 2 : A 2 includes nine base indexes B 2 ~ B29 evaluation atrix is:.6.4.3.5..2.. U ( A 2 ) =.8.2.6.3.2.6..3.4.4....2...7...2.5.2.......2...4.3..2.. According to the th row ( = ~ 9) of U A 2 distinguishable weights of B 2 are obtained and the distinguishable weight vector is: α ( 2 ).925.289.62.77.35.52.83.885) 2 In Table I iportance weight vector of B 2 ~ B 29 is given: β ( A 2 ) = (.9.8.6.6.3..4.7.7) 3 Calculate the K th coparable value of =,2 L9 and obtain the coparable value B 2 atrix N ( A 2 ) of A 2 :.8.22 N( A 2 ) =.84.68.29.6.54.37.5.4.46.34.88.42.6.5.54.7..29.56.3.8.9.42.7 4 According to N ( A 2 ), calculate the K th coparable su of A 2 and obtain the coparable su vector: M ( A 2 ) = (.445.38.24.69.7) 5 According to M ( A 2 ), calculate the ebership vector ( A 2 ) of A 2 : ( A 2 ) = (.422.3446.843.62.67) 3) Calculating the ebership vector of anageent security A 3 A 3 includes eight base indexes B 3 ~ B38 evaluation atrix is: U A 3 =.9.4.2.6.9..3.6..5.4.4...3.3...3...2.4.......6.. According to the th row ( = ~ 8)......2.. of U A 3 distinguishable weights of B 3 are obtained and the distinguishable weight vector is: α A 3 = (.29.84.42.3).56.465.29.3 2 In Table I iportance weight vector of B 3 ~ B 38 is given: β ( A 3 ) = (.8.2.9.2...7. ) 29 ACADEMY PUBLISHER
33 JOURNAL OF NETWORKS, VOL. 4, NO. 5, JULY 29 3 Calculate the K th coparable value of =,2 L9 and obtain the coparable value B 3 atrix N ( A 3 ) of A 3 : N A 3 =.45.5.2.5.99.42.73.6.63.39.7.22.42.37.3.29.2.55.2..62.2 4 According to N ( A 3 ), calculate the K th coparable su of A 3 and obtain the coparable su vector: M ( A 3 ) = (.634.288.3.72.2) 5 According to M ( A 3 ), calculate the ebership vector ( A 3 ) of A 2 : ( A 3 ) = (.5536.252.37.626.8) 4) Networ security coprehensive evaluation atrix U ( S ) Fro the above three steps, we can get networ U S : security coprehensive evaluation atrix U = ( S ) ( A ) = ( A2 ) ( A3 ).463.387.422.5536.3446.252.356.843.37.43.62.626.67.8 In Table I iportance weight vector of B ~ B3 is given: β ( S) = (.33.46.2 ) So we can calculate the final ebership vector S of the goal S : ( S ) = ( ( S ),..., 5 ( S )) = (.4594.3427.492.424.63) C. Networ Security Recognition Because the classification of networ security evaluation grades is orderly, that is, C is superior to C +, so we apply confidence recognition rule to deterine the grade of networ security. λ λ >.7 is the confidence, calculate Let K in t ( S ) λ, 5. = t= And udge that S belongs the K th grade, of which t S t= the confidence degree is no lower than. In the exaple, according to the final ebership vector ( S ), we can udge that S belongs the C 2 (good), with the confidence degree 8.2%(.4594+.3427=.82). In the Ref. [] final ebership vector of networ security evaluation is: ( S ) = ( ( S ),..., 5 ( S )) = (.3827.34794.8796.6956.84) The udgent result is the sae C 2 (good), but the confidence degree is only 73.64% (.3827+.34794 =.7364). V. CONCLUSIONS The transforation of ebership degree is the ey coputation of fuzzy evaluation for ulti-indexes fuzzy decision-aing, but the existing transforation ethods have soe questions. The paper analyzes the reasons of the questions, obtains the solving ethod, and at last builds the M (, 2, 3) odel without the interference of redundant data, which is different fro M (, + ) and is a nonlinear odel. M (, 2, 3) provides the general ethod for ebership transforation of ulti indexes decisionaing in application fields. The theory value is that it provides transforation ethod which is coply to logics to realize the transforation universe fuzzy set to universe fuzzy set in fuzzy logical syste. Fro index ebership degree of base level, after obtain one index ebership degree vector in adacent upper level by M (,2,3 ), thus, by the sae coputation, obtaining ebership degree vector of top level. Because of noralization of coputation, M (,2,3 ) is suitable for ebership transforation which contains ulti-levels, ulti-indexes, large data. Security is a very difficult topic. Everyone has a different idea of what security ' is, and what levels of ris are acceptable. The ey for building a secure networ is to tae preventive easures to solve networ security probles and iniize various networ security threats, so it is essential to evaluate networ syste security in advance. The paper applies the new algorith in the fuzzy evaluation on networ security and exaple results also prove that M (, 2, 3) can iprove the evaluation accuracy and credibility. REFERENCES [] Zhang Xihai., The Application of Fuzzy Coprehensive Evaluation for Networ Security Evaluation, Naning University of Technology and Engineering, 26. [2] Zadeh L.A., "Fuzzy sets", Inforation and Control, 965(8), pp.338 353. 29 ACADEMY PUBLISHER
JOURNAL OF NETWORKS, VOL. 4, NO. 5, JULY 29 33 [3] LI Feng qi, XIE Jun, LI,Yao, New Methods of Fitting the Mebership Function of Oceanic Water Masses, Periodical of Ocean University of China, 24., pp.-9. [4] Jun Chen, Susan Bridges, Julia Hodges, Derivation of Mebership Functions for Fuzzy Variables Using Genetic Algoriths, 998. [5] http://en.wiipedia.org/wii/mebership_function_(athe atics) [6] QIN Shou-ang et, The Theory and Application of Coprehensive Evaluation, Beiing: electronic industry publishing house, 23, pp.24. [7] WANG You-qiang, CHEN Shun-qing, HUANG Bing-xi, The Reliability Study of Colliery Machine Coponents, Colliery Machine, 24, (2), pp.53-55. [8] LV Ying-zhao, HE Shuan-hai, The Fuzzy Reliability Evaluation for Defect Reinforced Concrete on Bridges, Journal of Traffic and Transportation Engineering, 25, 5(4), pp.58-62. [9] HU Sheng-wu, LI Chang-chun, WANG Xin-zhou, GIS Quality Coprehensive Assessent Based on Multi-level Fuzzy Evaluation, Journal of Yangtze River Scientific research, 25, 22(3), pp.2-23. [] ZHENG Xian-bin, CHEN Guo-ing, Fuzzy Coprehensive Evaluation Research Based on FTA Oil and Gas Transport Vessel, Systes Engineering Theory & Practice, 25, (2), pp.39-44. [] LI Hai-ling, LIU Ke-ian,LI Qian, The Research of Fuzzy Coprehensive Evaluation for Proect Ris, Journal of XI Hua University, 24, 24(6), pp.78-8. [2] HUANG Yu-un, CUI Xin-yuan, JIA Sa-sa, Fuzzy Coprehensive Evaluation Research for Proect, Journal ZHE Jiang transportation college, 25,6(4), pp.-5. [3] HUANG Guang-long, YU Zhong-hua, Wu Zhao-tong, Subect-Obect Coprehensive Evaluation Based on Evidence Reasoning and Rough Set Theory, China Mechanical Engineering, 2, 2(8), pp:93-934. [4] GUO Jie, HU Mei-xin, The Iproveent of Fuzzy Evaluation Research for Proect Ris, Industrial Engineering Journal, 27, (3), pp.86-9. [5] Jianwei Han, Micheline, Data Mining: Concepts and Techniques, CA: Morgan Kaufan Publisher, 2. [6] JIA Lin, LI Ming, The Losing Model of Custoers of Teleco Based on Data Mining, Coputer Engineering and Aapplication, 24, pp.85-87. [7] YANG Wen-xian, REN Xing-in, QIN Wei-yang, Research of Coplex Equipent Fault Diagnosis Method, Journal of Vibration Engineering, 2, 3(5), pp. 48-5. [8] Zhang Weiqun., A Review and Forecast of the Study and Application of Data Digging, Statistics & Inforation Foru, 24, 9():,pp. 95-96. [9] Gao Yilong, Data Mining and Its Application in Engineering Diagnosis, Xi'an Jiaotong University, 2. [2] Liu Kai-Di, Pang Yan-Jun, Hao Ji-Mei, The Iproved Algorith of Fuzzy Coprehensive Evaluation for Vendor Selection of Iron and Steel Enterprises,. Statistics and Decision, 28(6), pp.7-73. Hua. Jiang, born in 977--9, Handan, Hebei Province, China. In March, 26, graduated fro Hebei University of Engineering and obtained postgraduate qualifications. Main research fields: networ security, inforation anageent, supply chain anageent. Now wors in Inforation Manageent Departent, Econoics and Manageent School, Hebei University of Engineering, Lecturer. Mainly published articles: Jiang, Hua, Study on obile E-coerce security payent syste, Proceedings of the International Syposiu on Electronic Coerce and Security, ISECS 28, Aug 3-5 28, pp.754-757; Jiang Hua, Ruan Junhu, Analysis of Influencing Factors on Perforance Measureent of the Supply Chain Based on SCOR-odel and AHM, IEEE/SOLI'28; Beiing, China October 2-5, 28, pp.24-246; Xiaofeng Zhao, Hua Jiang, Liyan Jiao, A Data Fusion Based Intrusion Detection Model, International Syposiu on Education and Coputer Science (ECS29), 7-8 March, 29 Wuhan, Hubei, China (in press). Current research interests: anageent optiization and scientific decision-aing. Junhu. Ruan, born in 983--2, Zhouou, Henan Province, China. Postgraduate of Hebei University of Engineering. Main research fields: logistic and supply chain anageent, scientific evaluation and prediction. Published article: Jiang Hua, Ruan Junhu, Analysis of Influencing Factors on Perforance Measureent of the Supply Chain Based on SCOR-odel and AHM, IEEE/SOLI'28; Beiing, China October 2-5, 28, pp.24-246. Current research interests: scientific evaluation and cobined forecasting. 29 ACADEMY PUBLISHER