2016/2/4 On a new grey clustering evaluation model Sifeng Liu et al Present by Naiming Xie E-mail:sifeng.liu@dmu.ac.uk; sfliu@nuaa.edu.cn Institute for Grey System Studies This work was supported by a project of Marie Curie International Incoming Fellowship within the 7th European Community Framework Programme entitled “Grey Systems and Its Application to Data Mining and Decision Support”(Grant No. FP7PIIF-GA-2013-629051), and a project of the Leverhulme Trust International Network entitled “Grey Systems and Its Applications” (IN-2014-020) 1 2016/2/4 Outlines 1 Introduction 2 The three kind of whitenization functions 3 The mixed center-point triangular whitenization functions 4 Steps of the grey clustering evaluation model using mixed center-point triangular whitenization functions 5 An Example 1 Introduction Grey clustering evaluation models, a kind of uncertain clustering evaluation model which using whitenization function is more suitable to be used to solve the problem of clustering evaluation with poor information, are used widely for uncertain systems analysis. 2 2016/2/4 1 Introduction Professor Deng Julong proposed the grey variable clustering model in 1986 using whitenization functions which is similar to the membership function (L.A. Zadeh, 1965) or the probability density function. Xiao Xinping(1997),Dong Fenyi(2010) and others are improved and optimized grey cluster evaluation models from different perspectives. Zhang Qishan studied measurement problem of Grey Characteristics of Grey Clustering Result(Q.S. Zhang, 2002). 1 Introduction The most used grey clustering model 1)The grey variable weight clustering model(J.L. Deng, 1986) ; 2)The grey fixed weight clustering evaluation model(S.F. Liu, 1993); 3)The grey cluster evaluation model using endpoint triangular whitenization functions(S.F. Liu et al, 1993),; 4)The grey cluster evaluation model using centerpoint triangular whitenization functions(S.F. Liu, N.M. Xie, 2011) etc. 3 2016/2/4 1 Introduction Grey variable weight clustering model is applicable to the problems with such criteria that have the same meanings and dimensions. When the criteria for clustering with different meanings, dimensions, grey fixed weight clustering evaluation model and grey cluster evaluation model using triangular whitenization function are suitable. 1 Introduction Compared with grey variable weight clustering model and grey fixed weight clustering model, Grey clustering evaluation model using end-point triangular whitenization functions suitable for the situation that all grey boundary is clear, but the most likely points belonging to each grey class are unknown; grey clustering evaluation model using center-point triangular whitenization functionsis suitable for those problems that is easier to judge the most likely points belonging to each grey class, but the grey boundary is not clear. 4 2016/2/4 1 Introduction S.F.Liu, B.J.Li et al presented a mixed whitenization function in 1998, but the mixed model hadn’t drew into the center-point triangular whitenization functions for 17 years. 2 Three kind of whitenization functions 1) The lower measure whitenization function f jk [ , , x jk ( 3 ), x jk ( 4 )] 0 f jk (x) 1 k xj (4) x xk (4) xk (3) j j x[0, xkj (4)] x[0, xkj (3)] x[xkj (3), xkj (4)] 5 2016/2/4 2 Three kind of whitenization functions 2) The triangular (moderate measure) whitenization function f k j [ x kj (1), x kj ( 2 ), , x kj ( 4 )] 0 k x xj (1) xk (2) xk (1) j k f j (x) j xk (4) x j xkj (4) xkj (2) x [xkj (1), xkj (4)] x [xkj (1), xkj (2)] x [xkj (2), xkj (4)] 2 Three kind of whitenization functions 3) The upper measure whitenization function f k j [ x kj (1) , x kj ( 2 ) , , ] 0, xxkj(1) k xxj (1) k fj (x) k , x[xkj(1),xkj(2)] k xj (2)xj (1) k 1, x x j (2) 6 2016/2/4 3 The mixed center-point triangular whitenization functions Assume that the turning point of grey class 1, grey 1 s class s are j,j and the center-point 2 j , 3 j , , s 1 j , s 1}) of grey class k(k {2,3, respectively. 3 The mixed center-point triangular whitenization functions For grey class 1 and grey class s, we take lower 1 1 2 measure whitenization function f j [ , , j , j ] , and s s1 s upper measure whitenization function f j [j ,j ,,] as corresponding whitenization function, and for , s 1}) , we take triangular grey class k ( k {2,3, whitenization function fjk(kj 1,kj ,,kj 1 ) as corresponding whitenization function. 7 2016/2/4 3 The mixed center-point triangular whitenization functions y y f j1(x) y f j 2 (x) y f j k (x) aj j1 j 2 j3 jk1j k jk1 y f j s1(x) y f j (x) s 1 o j s2 j s1 js bj x 图 6.4.2 中心点混合三角白化权函数示意图 4 Steps of the new grey clustering evaluation model Steps of the new grey clustering evaluation model using mixed center-point triangular whitenization functions as follows Step 1: Determine the turning point of grey class 1, grey class s 1j,sj and the center-point 2 j , 3 j , , s 1 j , s 1}) of grey class k(k {2,3, respectively according to the evaluation requirements and the grey class number s. 8 2016/2/4 4 Steps of the new grey clustering evaluation model Step 2: Setting the whitenization function of grey class 1, 2, …,s f j1[ , , 1j , 2j ] , s 1}) fjk(kj 1,kj ,,kj 1 ) k ( k {2,3, f js[sj1,sj ,,] Step 3: Determine the clustering weight for each index w , j 1, 2 , , m j 4 Steps of the new grey clustering evaluation model Step 4: Compute clustering coefficient i of object i regarding to grey class k k k i m j 1 f k j ( x ij ) w j ik } ik , determine the Step 5: According to max{ 1ks object i belonging to grey class k * Step 6:If there are several objects belonging to the same grey class, then sort them according grey clustering coefficient or decision coefficient with synthetic measure and principle of maximum value. 9 2016/2/4 5 An Example The evaluation of project for discipline construction of an university There are 6 primary indicators to reflect the performance of a project for discipline construction based on extensive surveys, 1)faculty, 2)scientific research, 3)students cultivation, 4)disciplines platform construction, 5)conditions for construction and 6)academic communication. The corresponding weights are 0.21, 0.24, 0.23, 0.14, 0.1,and 0.08 respectively. 5 An Example Disciplines construction evaluation Academic communication Construction conditions Disciplines platform Students cultivation Scientific research Faculty Fig.3 Evaluation indicator system of project for discipline construction 10 2016/2/4 5 An Example All the evaluation scores of the 6 indexes of 41 projects for discipline construction are laid in the interval of [40, 100] The evaluation results are divided into four grey class of 1, 2, 3, 4 corresponding to class poor, moderate, good and excellent, respectively. Step 1: Determine the turning point of grey class 1 and class 4, then the center-point of class 2, class 3 4 1j 60 j2 70,3j 80 j 90 5 An Example Step 2: Setting the whitenization function of grey class 1,2,3,4 0 x [40,70] 1 x [40,60] f( x ) 1 j 70 x x [60,70] 70 60 0 x 70 3 f( j x) 80 70 90 x 90 80 x [ 70 ,90 ] x [ 70 ,80 ] x [80 ,90 ] 0 x 60 2 f( j x) 70 60 80 x 80 70 x [ 60 ,80 ] x [ 60 , 70 ] x [ 70 ,80 ] 0 x [80,100] x 80 f(x) x [80,90] 9080 x [90,100] 1 4 j 11 2016/2/4 5 An Example Step 3: The weight for each index are 0.21, 0.24, 0.23, 0.14, 0.1,and 0.08 respectively. Step 4: Compute clustering coefficient Indicator name Faculty value 81 Academic Scientific Students Disciplines Construction conditions communication research cultivation platform 87 92 78 74 53 5 An Example grey class x1 x2 x3 x4 x5 x6 i excellent good medium poor 0.1 0.9 0 0 0.7 0.3 0 0 1.0 0 0 0 0 0.8 0.2 0 0 0.4 0.6 0 0 0 0 1.0 0.419 0.413 0.088 0.080 k i4 0.419 i3 0 . 413 Step 5: max 1 k 4 i It also shows that the execution effect of the project is situated between grey class "excellent" and "good“. 12 2016/2/4 Thanks for Attention! 13