JOURNAL OF COMPUTING, VOLUME 3, ISSUE 8, AUGUST 2011, ISSN 2151-9617 HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/ WWW.JOURNALOFCOMPUTING.

ORG

18

Using Clustering and DEA for evaluation and ranking university majors
Peyman Gholami, Azadeh Bazleh*, Mahdis Salehi
Abstract Although all university majors are prominent, and the necessity of their presence is of no question, they might not have the same priority basis considering different resources and strategies that could be spotted for a country. Their priorities likely change as the time goes by; that is, different majors are desirable at different time. If the government is informed of which majors could tackle today existing problems of world and its country, it surely more esteems those majors. This paper considers the problem of clustering and ranking university majors in Iran. To do so, a model is presented to clarify the procedure. Eight different criteria are determined, and 177 existing university majors are compared on these criteria. First, by k-means algorithm, university majors are clustered based on similarities and differences. Then, by DEA algorithm, we rank central point of each one clusters for evaluation and calculate the efficiency each clusters groups. Index Terms Clustering, Data mining, Data Envelopment Analyse (DEA), K-means algorithm, Multi-criteria decision making, University major ranking problem

1 INTRODUCTION
NIVERCITY major choice is an important decision to make for anybody seeking professional/higher education. It is a decision that will inuence the way people look at the world around themselves (Porter & Umbach, 2006). The future occupation of people is closely related to their education. Given this importance, it is always of interest to nd the guidance in collaboration with making aforementioned choices about which major to select. It is known that students should draw on available resources to ultimately pick a path that is right for them (Boudarbat, 2008). Nowadays, due to the creation of numerous undergraduate majors, the need for having a more precise approach becomes each time more necessary. Besides individual reasons, governments could be another client of university major choice. They might look for a way to supply their professional labors as one of the most inuential factors in its national future. To manage this and to nd which majors are of more important in future, they require a systematic approach to have more deep view about majors. For example, they entail to know areas each major affects, how majors can affect, to what extent each major is inuential in a given area. Although all university majors are prominent, and the necessity of their presence is of no question, they might not have the same priority basis considering different strategies that could be spotted for a country. Their priorities likely change as the time goes by; that is, different majors are desirable at different time. If the government is informed of which majors could tackle today existing problems of world and its country, it surely more esteems those majors. By more investing on those majors or providing greater grants for those studying the majors, they intend to motivate more talented students to study these majors. Therefore, with reference to the given explanations, it is a handy contribution to construct a model for such a decision-making process. To this end, we dene eight diffrent main specialization groups. We rst group university majors based on their similarities and differences which are obtained by their magnitude of inuence on MSGs. The values of different major group can then be calculated and evaluated to provide useful decisional information for the government to rationally exploit resources. Among available grouping methods, data mining approaches have been attracted more attention. Given different data mining models, clustering is regarded as the art of systematically nding groups in a data set (Fayyad, Piatetsky-Shapiro, & Smyth, 1996). In this paper, to cluster the university majors, we utilize the k-means algorithm as the most widely used method that have shown many successes in different applications such as market segmentation, pattern recognition, information retrieval, and so forth (Cheung, 2003; Kuo, Ho, & Hu, 2002). Besides its high performance, it is a very popular approach for clustering because of its simplicity of implementation and fast execution. Ranking/ordering university majors is a multi-criteria problem; that is, different criteria should be taken into account. For example, one major might be very important for industrial setting while another one is to improve so cial culture. Armed with this, we apply the data enve Department of Industrial Engineering, Arak Branch, Islamic Azad Univer- lopment analyse (DEA) as a simple multi-criteria decision sity, Arak, Iran making (or MCDM) method for dealing with unstruc * Department of Industrial Engineering, Arak Branch, Islamic Azad Unitured, multi-attribute problems. versity, Arak, Iran Department of Industrial Engineering, Arak Branch, Islamic Azad UniverDEA is a non-parametric linear programming based sity, Arak, Iran technique for measuring the relative efficiency of a set of similar units, usually referred to as decision making
2011 Journal of Computing Press, NY, USA, ISSN 2151-9617

JOURNAL OF COMPUTING, VOLUME 3, ISSUE 8, AUGUST 2011, ISSN 2151-9617 HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/ WWW.JOURNALOFCOMPUTING.ORG

19

units (DMUs). Because of its successful application and case studies, DEA has gained too much attention and widespread use by business and academy researchers. Evaluation of data warehouse operations (Mannino, Hong, & Choi, 2008), selection of flexible manufacturing system (Liu, 2008), assessment of bank branch performance (Camanho & Dyson, 2005), examining bank efficiency (Chen, Skully, & Brown, 2005), analyzing firms financial statements (Edirisinghe & Zhang, 2007), measuring the efficiency of higher education(by solving only one LP) institutions (Johnes, 2006), solving facility layout design (FLD) problem (Ertay, Ruan, & Tuzkaya, 2006) and measuring the efficiency of organizational investments in information technology (Shafer & Byrd, 2000) are examples of using DEA in various areas. Looking into the literature, there is no paper published dealing with the major choice as a nationwide problem. They almost tackle the problem as an individual assistance model. These papers usually propose regression models that guide a student to know which major is the best choice regarding her/his personal conditions, characteristics and interests (Porter & Umbach, 2006; Boudarbat, 2008; Berger, 1988; Crampton, Walstrom, & Schambach, 2006). As far as we reviewed, this paper is the rst work exploring this problem as a nationwide one, and cluster university majors using a data mining method called k-means. Moreover, university majors are ranked by a MCDM method, called DEA algorithm. The rest of the paper is organized as follows. Section 2 clusters the university majors. Section 3 presents the conceptual model of university majors ranking. Section 4 applies the DEA algorithm to order group university majors. Section 5 concludes the paper.

2 UNIVERCITY MAJOR RANKING MODEL

This section presents a conceptual model to describe the decision-making procedure of university major clustering and ranking. In fact, we employ a ow chart (FC) model to show whole procedure. This diagram is to clarify each step of whole procedure regardless of their details. Fig. 1 presents the FC model. The procedure could be divided into three main phases: Data gathering, Data preparation, and Decision making. In the rst phase, the list of existing university majors is solicited from Iranian Ministry of Science Research and Technology. University majors in Iran are presented in ve main groups each of which covers an educational background from high schools. These ve groups are: (1) Fine art, (2) Mathematics and Physics, (3) Empirical Sciences, (4) Human Sciences, (5) Foreign Languages. Finally, 177 university majors presented in Iran are identied. Then, MSGs are determined. Doing so, this paper intends to consider eight highlighted main specialization groups with due considerations to Irans own attributes and special areas are needed in order to ease
Fig. 1. The proposed Flow Chat (FC) model

2011 Journal of Computing Press, NY, USA, ISSN 2151-9617

JOURNAL OF COMPUTING, VOLUME 3, ISSUE 8, AUGUST 2011, ISSN 2151-9617 HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/ WWW.JOURNALOFCOMPUTING.ORG

20

TABLE 1 THE RESULTS OF UNIVERSITY MAJOR RANKING

2011 Journal of Computing Press, NY, USA, ISSN 2151-9617

JOURNAL OF COMPUTING, VOLUME 3, ISSUE 8, AUGUST 2011, ISSN 2151-9617 HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/ WWW.JOURNALOFCOMPUTING.ORG

21

TABLE 2 THE CENTRIODS OF UNIVERSITY MAJOR CLUSTERS IN EACH MSG IN PERCENTAGE

Cluster MSG Financial / Economical 44 59 11 16 7 18 29 6 41 4

Social/ Religious 28 9 14 54 37 5 2 20 59 2

Industry 18 70 0 4 4 18 23 4 19 3

Politics 4 6 0 5 41 1 4 10 50 2

Services 11 41 47 25 8 20 16 8 24 6

Therapy / Health 2 18 4 8 1 8 44 2 6 1

Agriculture 5 21 58 37 2 10 35 5 5 3

1 2 3 4 5 6 7 8 9 10

Environment/ Natural Resources 10 16 15 27 1 10 54 3 6 3

the design process of sustainable development. These eight MSGs were extracted after a review of the literature of the problem and the reports published by the local government for achieving sustainable development, and the validity and reliability of these MSGs have been veried and conrmed by a number of structured interviews. At this stage, additional rules and constraints taken from Irans strategies and views should be considered as well. Finally, the following eight MSGs are considered as decision criteria: 1. Financial/Economical group 2. Social/Religious group 3. Industrial group 4. Political group 5. Service group 6. Agricultural group 7. Therapy/Health group 8. Environmental/Natural source group In the second phase, regarding the data gathered in previous phase, two suitable questionnaires are designed. The rst one is to compare university majors on their magnitude of inuence on above-mentioned MSGs. The second one is to compare the importance/weight of each MSG for today Iran. The questionnaires are sent to several experts (64 experts) whose denition is set in this research as follows: an expert is a person who has at least a Master of Science degree in one of the ofcial university majors along with at least a three-year working experience in his/her specialization. When data are collected, they are prepared. If all the requirements are met, the third phase starts. First, we employ one of the well-known data mining approaches to cluster university majors based on their similarities and differences on the results. This step is explained in more details in Section 3. Then, we rank university majors by means of a MCDM algorithm. There are two options to employ: (1) multi-objective decision maing (or MODM), or (2) multi-attribute decision-making (or MADM) approaches. MODM models are those searching a continuous/integer space to nd optimal solutions. The most commonly used type of these models is linear programming. Although the MODM model is not a good choice for the problem studied here, we formulate the university major ranking problem in the form of a linear programming model. The following notications are dened.

n m i j b ij

the number of majors (n = 177) the number of MSGs (m = 8) index for majors; i = {1, 2, . . . , n} index for MSGs; j = {1, 2, . . . , m} the magnitude of inuence of major i on MSG j

The decision variables are as follows. X i continuous variable for the importance rate of major i The proposed model is given below: Maximize Subject to:

Eq. (1) calculates the objective function which is the summation of multiplying the importance rate given to each major and its inuence on each of all eight MSGs. Constraint set (2) limits the total importance of all the majors to be equal to 1. Constraint set (3) denes the decision variables. Since the problem of ranking university majors is not a continuous problem, the MODM model is not the best choice. Our purpose to present the model is to mathematically characterize the problem. A MADM model could be more effective. Among the MADM approaches, DEA has shown many successful applications in such ranking problems. Therefore, we have been thinking of ranking the university majors by DEA algorithm. The details and results are presented in Section 4.

3 UNIVERSITY MAJOR CLUSTERING PROBLEM (UMCP) IN IRAN

3.1 THE BACKGROUND OF CLUSTERING AND K-MEANS ALGORITHM
In today world, data are considered as one of the most valuable assets. With the current dramatic increase in magnitude of available data and also their low cost storage, it became interesting to discover knowledge in these data. Therefore, the importance of how to effectively process and use data more and more soars. This calls for

2011 Journal of Computing Press, NY, USA, ISSN 2151-9617

JOURNAL OF COMPUTING, VOLUME 3, ISSUE 8, AUGUST 2011, ISSN 2151-9617 HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/ WWW.JOURNALOFCOMPUTING.ORG

22

new techniques to help analyze, understand the huge amounts of stored data (Liao & Chen, 2004). Among the new techniques developed, data mining is the non trivial extraction of hidden and potentially useful information from large sets of data. In other words, data mining is the process of discovering signicant knowledge, such as patterns, associations, changes, anomalies and signicant structures from large amounts of data stored in databases, data warehouses, or other information repositories (Liao, Chen, & Wu, 2008). In the literature, there are many data mining models such as classication, estimation, predictive modeling, clustering, afnity grouping or association rules, description and visualization, as well as sequential modeling. Clustering is a widely used technique, whose goal is to provide insight into the data by partitioning the data (objects) into disjoint and homogeneous groups (clusters) of objects, such that objects in a cluster are more similar to each other than to objects in other clusters. According to Boutsinas and Gnardellis (2002), clustering algorithms have been frequently studied in various elds including machine learning, neural networks and statistics, among others (Fensel, 2001; Corcho, Lopez, & Perez, 2003; Davies & Fensel, 2003). The k-means algorithm, rst proposed by MacQueen (1967), is the most popular partition-clustering method that has attracted great interest in the literature. The goal of the k-means algorithm is to partition the objects into k clusters so that the within-group similarity is maximized. The procedure of k-means methods could be described as follows. 1. Place k points into the space represented by the objects that are being clustered. These points represent intial group centroids. 2. Assign each object to the group that has the closest centroid. 3. When all objects have been assigned, recalculate the positions of the k centroids. 4. Repeat Steps 2 and 3 until the centroids no longer move. This produces a separation of the objects into groups from which the metric to be minimized can be calculated.

distance measure in k-means method (Huang, Chang, & Wu, 2009). That is, the distance between the major UM i and the centeriod C l is obtained by the following formula:

Again, each major is assigned to the nearest cluster, and the new centeriod dimension j of cluster l is the arithmetic mean of UM ij for the majors belonging to the cluster l. This procedure iterates until no new cluster is obtained when majors are re-assigned. To run the procedure, the algorithm coded in MATLAB 7. The results of UMCP and the nal centriods of the 10 clusters are presented in Tables 1 and 2, respectively. Centeriods of eachcluster are its average magnitude of inuence on each MSG. Cluster 1 includes the majors concerning more on nancial and economical MSG, and slightly on social and religious MSG. Cluster 2 consists of engineering majors; therefore, they clearly focus on the industrial MSG. Majors in Cluster 3 are those relating to individual therapy and health, whereas we could see majors of Cluster 4 are those focusing on public health. Cluster 5 covers majors training social related courses such as social, political, religious and military affairs. Cluster 6 involves majors providing services for civilians. Cluster 7 includes majors relating to Agriculture and Natural Resources MSG. Cluster 8 consists of majors that their aspects of social services are more inuential than the other aspects. Cluster 9 involves majors that could almost affect all the eight criteria; although they are more important on economical, social and religious, political, and service criteria. Cluster 10 apparently covers majors that have been given lower values by the involved decision makers. Based on results collected, they might be comparatively less inuential.

4 UNIVERSITY MAJOR RANKING PROBLEM (UMRP) IN IRAN

4.1 THE BACKGROUND OF DATA ENVELOPMENT ANALYSE (DEA)
DEA is a technique for deciding the relative efficiency of a Decision Making Unit (DMU) by comparing it with linear combinations of other DMUs engaged in providing the same outputs from the same inputs [23]. Charnes, Cooper, and Rhodes developed data envelopment analysis (DEA) to evaluate the efficiency of decision making units (DMUs) through identifying the efficiency frontier and comparing each DMU with the frontier [20]. Since DEA is able to estimate efficiency with minimal prior assumptions [28], it has a comparative advantage to approaches that require a priori assumptions, such as standard forms of statistical regression analysis [4]. During the past thirty years, various DEA extensions and models have been developed and established themselves as powerful analytical tools [23]. The original DEA model presented by Charnes, Cooper, and Rhodes [20] is called CCR ratio model, which uses the ratio of outputs to inputs to measure the

3.2 THE APPLICATIONS OF K-MEANS FOR UNIVERSITY M AJOR CLUSTERING PROBLEM

This study employs k-means in cluster analysis and partitions 177 university majors into 10 clusters. Let X = {UM 1 ; UM 2 ; . . . ; UM n = 177} be the set of majors to be clustered where UM i indicates the university major i. Using the vector-space model, each major is measured with respect to a set of m = 8 initial attributes A 1 ; A 2 ; . . . ; A 8 where A j indicates MSG j. Each major is therefore dscribed by a 8-dimensional vector UMi = {UM i , 1 ; . . . , UM i , 8 } where UM i,j is UM i s magnitude of inuence on MSG j; UM ij R; 1 i 177; 1 j 8. The initial 10 centriods C l , l = {1, 2,. . . , 10} are randomly selected. Let C lj indicates the value of dimension j of centriod l. Then, the distance between each major and centeriod is calculated using Euclidean distance as the most commonly used

2011 Journal of Computing Press, NY, USA, ISSN 2151-9617

JOURNAL OF COMPUTING, VOLUME 3, ISSUE 8, AUGUST 2011, ISSN 2151-9617 HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/ WWW.JOURNALOFCOMPUTING.ORG

23

efficiency of DMUs. Assume that there are n DMUs with m inputs to produce s outputs. X ij and y rj represent the amount of input i and output r for DMU j (j = 1, 2, . . .,n), respectively. Then the ratio-form of DEA can be represented as:

TABLE 3 THE EFFICIENCY OF UNIVERSITY MAJOR CLUSTERS

where the u r s and the v i s are the variables and the y and x io s are the observed output and input values of the DMU to be evaluated (i.e., DMU o ), respectively [23]. The equivalent linear programming problem using the CharnesCooper transformation is [23]:
ro s

efficient when its input values are all enlarged by 50%. If the cluster is inefficient then the AP-value is equal to the efficiency value. In this paper, the CCR and BCC models are utilized.

4.2 THE APPLICATION OF DEA FOR UMRP

c

Banker, Charnes, and Cooper introduced the BCC model to the CCR model [5]. by adding a constraint These models can be solved using the simplex method for each DMU. DMUs with value of 1 are efficient and others are inefficient. Nakhaeizadeh and Schnabl proposed to use DEA approach in data mining algorithms selection [26]. They argued that in order to make an objective evaluation of data mining algorithms that all the available positive and negative properties of algorithms are important and DEA models are able to take both aspects into consideration. Positive and negative properties of data mining algorithms can be considered as output and input components in DEA, respectively. For example, the overall accuracy rate of a classification algorithm is an output component and the computation time of an algorithm is an input component. Using existing DEA models, it is possible to give a comprehensive evaluation of data mining algorithms. In this study, input components are equal . Output components include the scors that each centroid gained from criteria (MSG). After solving LP, the clusters with h r = 1 (100%) are efficient clusters. The other clusters do not belong to the efficiency frontier and remain outside of it. As already mentioned, the definition of efficiency is more general than intrestingness as suggested by Fayyad et al. (1996). For ranking the clusters, one can use the approach suggested by Andersen and Petersen (1993) (AP-model). They use a criterion that we call it the AP-value. In inputoriented models the AP-value measures how much an efficient cluster can radially enlarge its input-levels while remining still efficienct (output-oriented is analogous).
R R

In this section using DEA the central points of every cluster will be consider as DMUs and then with applying of 8 criteria that already mentioned we are able to evaluate the efficiency of each cluster that the results summarized in table 3. As you see in the table3 the efficiency of more than 1 DMUs are equal to 1. Therefore, using A-P model we calculate the efficiency of them again and ultimately we will able to final ranking of these clusters. The results represented in table 4. As shown in table 4 the first rank is cluster 2.

TABLE4 THE EFFICIENCY OF UNIVERSITY MAJOR CLUSTERS OBTAINED FROM A-P MODEL

5 CONCLUSION
This paper dealt with university majors ranking problem. The UMRP is an important problem since university majors might not have the same priority basis with due considerations to different resources and strategies that a country has, they are all eminent though. The UMRP is a dynamic problem; therefore, a general model is needed to

For example, for an input-oriented method an APvalue equal to 1.5 means that the algorithm remains still

2011 Journal of Computing Press, NY, USA, ISSN 2151-9617

JOURNAL OF COMPUTING, VOLUME 3, ISSUE 8, AUGUST 2011, ISSN 2151-9617 HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/ WWW.JOURNALOFCOMPUTING.ORG

24

clarify the whole procedure. In this case, we employed a Flow Chart model which has three phases: Data gathering, Data preparation, and Decision making. In the rst two phases, all the data needed for the third phase are collected and tested for the necessary requirements. In the third phase, the all the majors are clustered according to their similarity and differences by k-means algorithm. Since UMRP is an MADM problem, we employ an application of DEA algorithm to rank university majors. As a direction for future research, one might work on application of other multi objective decision making procedure.

[17] Kuo, R. J., Ho, L. M., & Hu, C. M. (2002). Integration of self-organizing feature map and K-means algorithm for market segmentation. Computers and Operations Research, 29(11), 14751493. [18] Liao, S. H., & Chen, Y. J. (2004). Mining customer knowledge for electronic catalo marketing. Expert Systems with Applications, 27, 521532. [19] Liao, S. H., Chen, C. M., & Wu, C. H. (2008). Mining customer knowledge for product line and brand extension in retailing. Expert Systems with Applications, 35(3), 17631776. [20] Lipovetsky, S., & Conklin, W. M. (2002). Decision aiding. Robust estimation of priorities in the AHP. European Journal of Operational Research, 137, 110122. [21] Liu, S. T. (2008). A fuzzy DEA/AR approach to the selection of flexible manufacturing systems. Computers & Industrial Engineering, 54, 6676. [22] Kablan, M. M. (2004). Decision support for energy conservation promotion: Ananalytic hierarchy process approach. Energy Policy, 32, 1151 1158. [23] Koksalan. M, C. Tuncer, A DEA-based approach to ranking multicriteria alternatives, International Journal of Information Technology and Decision Making 8 (1) (2009) 2954. [24] MacQueen, J. B. (1967). Some methods for classication and analysis of multivariate observations. Proceedings of 5-th Berkeley symposium on mathematical statistics and probability (vol. 1, pp. 281297). Berkeley: University of California Press. [25] Mannino, M., Hong, S. N., & Choi, I. J. (2008). Efficiency evaluation of data warehouse operations. Decision Support Systems, 44, 883898. [26] Nakhaeizadeh G., A. Schnabl, Development of multi-criteria metrics for evaluation of data mining algorithms, in: Proceeding of the Third International Conference on Knowledge Discovery and Data Mining (KDD97), Newport Beach, California, August 1417, 1997, pp.3742. [27] Porter, S. R., & Umbach, P. D. (2006). College major choice: An analysis of personenvironment t. Research in Higher Education, 47(4), 429 449. [28] Shafer, S. M., & Byrd, T. A. (2000). A framework for measuring the efficiency of organizational investments in information technology using data envelopment analysis. Omega, 28, 125 141. [29] Steuer, R. E. (2003). Multiple criteria decision making combined with nance: A categorized bibliographic study. European Journal of Operational Research, 150, 496515.

REFERENCES
[1] [2] Berger, M. C. (1988). Predicted future earnings and choice of college major. Industrial and Labor Relations Review, 41, 418429. Boudarbat, B. (2008). Job-search strategies and the unemployment of university graduates in Morocco. International Research Journal of Finance and Economics, 14, 1533. Boutsinas, B., & Gnardellis, T. (2002). On distributing the clustering process. Pattern Recognition Letters, 23, 9991008. Camanho, A. S., & Dyson, R. G. (2005). Cost efficiency measurement with price uncertainty: A DEA application to bank branch assessments. European Journal of Operational Research, 161, 432446. Chen, X., Skully, M., & Brown, K. (2005). Banking efficiency in China: Application of DEA to pre- and post-deregulation eras: 19932000. China Economic Review, 16, 229245.

[3] [4]

[5]

[6]

Chen, Y. L., & Cheng, L. C. (2009). Mining maximum consensus sequences from group ranking data. European Journal of Operational Research, 198, 241251.\ [7] Cheung, Y. M. (2003). K-means: A new generalized K-means clustering algorithm. Pattern Recognition Letters, 24, 28832893. [8] Corcho, O., Lopez, M. F., & Perez, A. G. (2003). Methodologies, tools and languages for building ontologies. Where is their meeting point? Data and Knowledge Engineering, 46, 4164. [9] Crampton, W. J., Walstrom, K. A., & Schambach, T. P. (2006). Factors inuencing major selection by college of business students. Information Systems, 7(1), 226230. [10] Davies, J., & Fensel, D. (2003). Toward the Semantic Web: Ontology driven knowledge management. John Wiley & Sons Ltd. [11] Edirisinghe, N. C. P., & Zhang, X. (2007). Generalized DEA model of fundamental analysis and its application to portfolio optimization. Journal of Banking & Finance, 31, 33113335. [12] Ertay, T., Ruan, D., & Tuzkaya, U. R. (2006). Integrating data envelopment analysis and analytic hierarchy for the facility layout design in manufacturing systems. Information Sciences, 176, 237262. [13] Fayyad, U. M., Piatetsky-Shapiro, G., & Smyth, P. (1996). Advances in knowledge discovery and data mining. Cambridge: AAAI Press/MIT Press. [14] Fensel, D. (2001). Ontologies: A silver bullet for knowledge management and electronic commerce. New York: Springer. [15] Huang, S. C., Chang, E. C., & Wu, H. H. (2009). A case study of applying data mining techniques in an outtters customer value analysis. Expert Systems with Applications, 36, 59095915. [16] Johnes, J. (2006). Measuring teaching efficiency in higher education: An applicationof data envelopment analysis to economics graduates from UK Universities. European Journal of Operational Research, 174, 443456.

2011 Journal of Computing Press, NY, USA, ISSN 2151-9617