Decision Support Systems 54 (2013) 1150–1160

Contents lists available at SciVerse ScienceDirect

Decision Support Systems

j ourna l homepage: www.e lsev ie r .com/ locate /dss

A comparative analysis of machine learning systems for measuring the impact ofknowledge management practices

Dursun Delen a,⁎, Halil Zaim b, Cemil Kuzey c, Selim Zaim d

a Department of Management Science and Information Systems, Spears School of Business, Oklahoma State University, United Statesb Department of Management, Fatih University, Buyukcekmece, Istanbul, 34500, Turkeyc Department of Management, Fatih University, Buyukcekmece, Istanbul, 34500, Turkeyd Department of Mechanical Engineering, Marmara University, Goztepe, Istanbul, 34722, Turkey

⁎ Corresponding author. Tel.: +1 918 594 8283; fax:E-mail addresses: dursun.delen@okstate.edu (D. Del

(H. Zaim), cemilkuzey@gmail.com (C. Kuzey), selim.zaim

0167-9236/$ – see front matter © 2012 Elsevier B.V. Allhttp://dx.doi.org/10.1016/j.dss.2012.10.040

a b s t r a c t

a r t i c l e i n f o

Article history:Received 29 May 2012Received in revised form 20 September 2012Accepted 28 October 2012Available online 1 November 2012

Keywords:Knowledge managementMachine learningPredictive modelingService industryImpact analysis

Knowledge management (KM) has recently emerged as a discrete area in the study of organizations andfrequently cited as an antecedent of organizational performance. This study aims at investigating the impactof KM practices on organizational performance of small and medium-sized enterprises (SME) in serviceindustry. Four popular machine learning techniques (i.e., neural networks, support vector machines, decisiontrees and logistic regression) along with statistical factor analysis (EFA and CFA) are used to developedpredictive and explanatory models. The data for this study is obtained from 277 SMEs operating in the serviceindustry within the greater metropolitan area of Istanbul in Turkey. The analyses indicated that there is astrong and positive relationship between the implementation level of KM practices and organizational per-formance related to KM. The paper summarizes the finding of the study and provides managerial implicationsto improve the organizational performance of SMEs through effective implementation of KM practices.

© 2012 Elsevier B.V. All rights reserved.

1. Introduction

As one of the contemporary management tools, knowledge manage-ment (KM) has been increasing in popularity of the tools/techniquesused by large organizations and multinational companies to gainsustainable competitive advantage in the long run. Despite the growinginterest and implementation initiatives, the concept of KM is stillevolving, and to date there is no unifying or overarching theoreticalframework that has been widely accepted.

While KMhas been frequently cited as an antecedent of organization-al performance, there is a paucity of empirical research regarding the im-pact of KM practices on organizational performance. This lack of interestis even more pronounced in the context of small and medium-sized en-terprises (SMEs). While implementation of KM practices in large sizefirms provides immense business opportunities in terms of achievingcost efficiency and gaining competitive advantage, there is less evidenceof small- andmedium-sized enterprises (SMEs) implementing KM prac-tices to capture similar benefits. The question then remains open “howwell KM practices fit with the SMEs”, which form the largest group ofbusiness establishments in both developed and emergingmarket econo-mies from the viewpoint of generating employment and economicgrowth [12]. They account for more than half of the employment andvalue added contributions in most countries [50]. Similar trend is also

+1 918 594 8281.en), halilzaim@fatih.edu.tr@marmara.edu.tr (S. Zaim).

rights reserved.

observed in Turkey where SMEs constitute more than 90% of the totalnumber of businesses and employ 61% of the workforce [53].

In view of the fact that the success of SMEs has a direct impact onthe national economy, this study aims to provide two main contribu-tions to SME research. First, based on a sample of SMEs operating intwo sub-sectors of textile industry within the greater metropolitanarea of Istanbul in Turkey, this study aims to examine the impact ofKM practices on the organizational performance of SMEs. Second,the machine learning approach, which has been gaining growinginterest in business research, is employed to identify the most impor-tant KM practices on organizational performance of SMEs.

The remainder of the paper is organized as follows. The nextsection provides a rather comprehensive review of the relevant liter-ature on KM practices. Research methodology is presented inSection 3. Data analysis, results and their implications are providedin Section 4. The paper concludes with Section 5 where a summaryof the findings along with future research directions are given.

2. Literature review

The field of KM has recently emerged as a new area of interest forboth academic and business circles. The review of the recent litera-ture reveals an increasing number of studies covering many differentfacets of KM [38]. Along with this growing interest, researchers pro-posed a large number of definitions of KM, most of which overlappingon common characteristics [33], while each emphasizing on a fewdistinct aspects of KM. Generally speaking, the existing studies in

1151D. Delen et al. / Decision Support Systems 54 (2013) 1150–1160

the field of KM have largely focused on three major streams [17]: thephilosophical nature of KM; the processes of knowledge management(i.e., generation, sharing and distribution of knowledge); and theinfrastructure of knowledge management in terms of technology andeffective management of knowledge and business practices. Zaim etal. [56] classify the infrastructure further into four areas: technology,organizational culture, organizational structure and intellectualcapital. Similarly they also identify four areas of processes for KM:knowledge generation and development; knowledge codificationand storage; knowledge transfer and sharing; and knowledge utiliza-tion. In the forthcoming section, we will develop the concept of KM inline with the categorization purported by Zaim et al. [56], and willsubsume both KM processes and related Knowledge ManagementInfrastructure under the general heading of KM practices.

2.1. KM practices

It has been argued that the effectiveness of KM depends on howthe generation of new knowledge is organized and how existingknowledge is transferred throughout the organization. Recent studieshave expressed considerable interest in knowledge sharing practices[24]. The benefits of knowledge transfer and sharing have also beendiscussed widely among the scholars and practitioners [48]. There-fore, one of the most important objectives of KM is to bring togetherintellectual resources and make them available across organizationalboundaries. It has been suggested that organizations often waste theirresources and lose a significant amount of money for repeating thesame mistakes, duplicating projects and being unaware of eachother's knowledge due to the lack of knowledge transfer and sharingthroughout the organization [44].

Knowledge transfer is not a unidirectional movement. Effectiveknowledge transfer is more than the movement of knowledge fromone location to another. Organizations can get significant learning expe-rience through knowledge transfer between units and people. It tends toimprove competency of both sides that transfer and share knowledge. Itis because knowledge does not leave the owner when it has been trans-ferred. As a result, the value of knowledge grows each time a transfertakes place and the key to value creation lies in how effective knowledgehas been transferred throughout the organization.

The role and importance of information and communication tech-nologies in knowledge transfer have been emphasized by manyscholars. Clearly, technological advances bring a vast number of newopportunities to transfer and share knowledge and expertisethroughout the organization within departments, plants, countriesand across national borders. These technologies have a strategic rolein knowledge sharing specifically for the geographically dispersedglobal organizations [2]. The effective use of technologies createsnew ways of knowledge transfer and hold promising solutions bothin transfer of explicit knowledge and tacit knowledge — in terms ofexperience and expertise [26]. In this respect, it is often mentionedthat technological infrastructure has a strategic importance in knowl-edge transfer not only within the organization but also among differ-ent organizations [57].

As a matter of fact, all healthy organizations generate knowledge.While they are interactingwith their environment, they absorb informa-tion, combine it with their experiences, values and internal rules, turn itinto knowledge, and take action based on it. Knowledge generation can beperformed inmanyways. The three of themainmodes among others areknowledge acquisition, knowledge generation within the firm andcollaborative knowledge generation. However, knowledge generationprocess is a set of activities for the conscious and intentional generationof knowledge under specific actions and initiatives firms undertake toincrease their stock of corporate knowledge [10].

Knowledge generation process does not necessitate new knowl-edge generation. In many circumstances, organizations may preferto acquire knowledge from other sources and adopt it for their own

use [4]. Knowledge acquisition can be used for knowledge creation,and if it is novel and useful for the organization, also be consideredas a part of knowledge generation. Organizations convert informationthey collect from internal and external sources into knowledgethrough their organizational learning process by combining it withtheir prior knowledge, experiences, values and organizational proce-dures [25]. Then, the knowledge becomes a part of their organization-al knowledge base. This obviously explains why the knowledgeacquired through these organizational processes is new and uniquefor that organization [29].

Knowledge is meaningful when it is codified, classified, put in a use-ful format and stored. Only then, it can be used by the right person, atthe right time and in the right way. Knowledge codification and storageis important not only for an effective use of knowledge but also for reus-ability of knowledge in case it is needed so that the knowledge in ques-tion can be internalized to the organization rather than the knower [39].Therefore, considering the organization's overall objectives and priori-ties, many studies have been concentrating on the classification andthe codification of knowledge based on its types and purposes [32],and on the storage of knowledge to let the employees be able to accessknowledge any time both at present and in the future. The codificationof knowledge also enables to stock the knowledge resources and to as-sess the potential of the organization. The most challenging feature ofknowledge codification is to extract it without losing its distinctiveproperties which makes it valuable [10].

Despite its importance, codifying and classifying knowledge is notthat simple since it relies heavily on what people know. Thus, organi-zational knowledge is hard to capture, clarify and express perfectlyfine considering the fact that it is dispersed and scattered throughoutthe organization. It is found in different locations, in peoples' minds,in various organizational processes, in corporate culture embeddedinto different artifacts and procedures and stored into differentmediums such as print, disks and optical media [5].

There is a distinction between tacit and explicit knowledge in thestorage of knowledge. Explicit knowledge can be easily collected,documented, stored and retrieved quite independently of any singleindividual through technological means and systems. On the otherhand, tacit knowledge resides in the minds of the employees andseizes a great deal of an organization's knowledge resources [14]. Ifthe organization's knowledge resources have been described as aniceberg, the explicit knowledge is the visible part of the icebergabove the surface, whereas the tacit knowledge includes the invisiblepart of the iceberg beneath the surface [23]. The codification of tacitknowledge unlike explicit ones is the most cumbersome activity inthe overall process because of its subjective and situational nature,and it is intimately tied to the knower's experience.

One of the most important and challenging aspects of KM is to en-hance the development of a collaborative, trustworthy, emphatic andhelpful organizational culture. The executives and scholars agree on theimportance of a knowledge-friendly culture for the success of KM[21,45]. It is because knowledge is a context-dependent social concept[30] and a large part of organizational knowledge is embodied in socialprocesses, institutional practices, traditions and values [6,15]. Therefore,nomatter how powerful the tools and functions of KM are, it is of no usewithout willing participants and a supportive social and cultural envi-ronment [28]. While the cultural resistance is generally cited as one ofthe most important barriers to an effective implementation of KM[48], it is still contemplated as the neglected or underestimated side ofKM practices. Therefore, it is strictly recommended for organizationsto place a special emphasis on the social and cultural issues for the suc-cessful implementation of KM practices [5].

The appropriate organizational structure and guidelines as well astechnical and non-technical expedients of which the organization hasdisposal constitute another building blocks of KM infrastructure [1].Nonetheless, there is no single appropriate organizational structure forKM. Some scholars suggest a radical re-design for KM [35], while others

HN3

HN2

HN1

OC

IP

KTS

TI

OS

KG

KU

KCS

Input Layer Hidden Layer Output layer

PERF

Fig. 1. Structure of the NN model.

1152 D. Delen et al. / Decision Support Systems 54 (2013) 1150–1160

think that it is not necessary. However, instead of highly centralized,control-based and rigid hierarchies, more flexible, decentralized andtrust-based organizational structures with empowered workers arehighly recommended in the KM literature [34,36].

One of the most important objectives of KM is to create value fromorganization's knowledge resources so that the knowledge held by thecompany can be transformed to fields of application and action [43].This implies the effective and efficient knowledge utilization for theorganization's competitive edge. For that reason, it has been argued thatthe success of KMpracticesmostly depends on howefficient and effectivethe knowledge has been used and the level of action based on it [52].

The KM literature clearly exposes that knowledge resources havebeen increasingly seen as an integral part of organizations' value cre-ating processes. In a similar vein, companies have become aware ofthe importance of intellectual capital of their own [19]. Intellectualcapital can be defined as ‘the sum of all the intellectual materials ofa company’ – knowledge, information, intellectual property includingtrademarks, patents and licenses, experience and integrity, personnelcompetencies, collective brainpower, etc. – that is captured and lever-aged to create value and that can be converted to wealth and profit[7,20,47]. Though there are a variety of different components thatconstitute intellectual capital, an increasingly popular classificationdivides intellectual assets into three categories: human capital, struc-tural capital and customer capital [46].

2.2. Performance

The main objective of KM performance evaluation is to increasethe effectiveness, efficiency and adaptability of KM efforts so as toadd more value to the overall performance of the organization [49].

Given the general rule about performance evaluation that perfor-mance improves through evaluation, it is reasonable to argue that mea-suring the outcomes and evaluating the contribution of KM practices areimportant to ensure the sustainability and success of KM efforts overtime. Without assembling the link between desired outcomes and KMpractices continuously and demonstrating tangible or quantifiable in-tangible results, it is not possible for the top management to keep oninvesting and for the workers to preserve their concentration and moti-vation [31,41]. Apparently, evaluation of KM-related organizational per-formance also shows to what extent the intellectual resources of a firmhave been utilized [16,37] as well as the degree of the conversion of theorganizational knowledge into improved performance [27].

3. Method

What follows is a brief introduction to the machine learning tech-niques (i.e., artificial neural networks, support vector machines, deci-sion trees and logistic regression) employed in this study aspredictive modeling tools. These four techniques are selected basedon their superior predictive performance and their popularity in therecently published research literature.

3.1. Neural network

Artificial neural networks (ANNs) are analytic techniquesmodeled on the learning processes of the human cognitive systemand the neurological functions of the brain. Recently, there has beena considerable interest in the development of artificial neural net-works for solving a wide range of problems from a variety of fields.Neural networks are distributed information processing systemscomposed of many simple computational elements interacting acrossweighted connections. Inspired by the architecture of the humanbrain, neural networks exhibit certain features such as the ability tolearn complex patterns of information and generalize the learned in-formation. Neural networks are simply parameterized non-linearfunctions that can be fitted to data for prediction purposes.

Artificial neural networks can be classified into several categoriesbased on supervised and unsupervised learning methods andfeed-forward and feedback recall architectures. A back propagationneural network (BPNN) uses a supervised learning method andfeed-forward architecture. A BPNN is one of the most frequently uti-lized neural network techniques for classification and prediction [22].

The main appeal of neural networks is their flexibility in approxi-mating a wide range of functional relationships between inputs andoutputs. Indeed, sufficiently complex neural networks are able to ap-proximate arbitrary functions arbitrarily well. One of the most inter-esting properties of neural networks is their ability to work andforecast even on the basis of incomplete, noisy, and fuzzy data. Fur-thermore, they do not require a priori hypothesis and do not imposeany functional form between inputs and outputs. For this reason, neu-ral networks are quite practical to use in the cases where knowledgeof the functional form relating inputs and outputs is lacking, or whena prior assumption about such a relationship should be avoided.

The success of the ANN models depends on properly selected pa-rameters such as the number of nodes (neurons) and layers, thenonlinear function used in the nodes, learning algorithm, learning pa-rameters (learning rate and momentum), initial weights of the inputsand layers, and the number of epochs (i.e., cycles) that the model isiterated. The structure of a typical ANN model consists of one inputlayer, one or more hidden layers and one output layer. A graphical de-piction of the specific ANN model (i.e., multi-layered perceptron withone hidden layer) used in this study is shown in Fig. 1.

In ANN methodology, the sample data often is divided into twomain sub-samples which are named as training and test sets. Duringthe training process, the neural network learns the relationship be-tween output and input criteria, while in the testing process, testset is used to assess the true predictive performance of the model.

3.2. SVM (support vector machine)

Support vectormachine (SVM), originally developedbyVapnik [51], isamong themost robust and accurate methods in datamining algorithms.Its theoretical foundation is derived from statistical learning theory. SVMcombines the statistical methods and machine learning methods.

SVM is a supervised learning method that generates input–outputmapping functions from a set of training data. Basically SVM learnsfrom observations. There is an input space and output space and atraining set. The nature of the output space decides the learningtype such as type of binary or multiple classification problems.

Develop the survey instruments(regarding the knowledge managementcriteria and organizational performance).

Select the survey sample,collect the data,examine and pre-process the data for

completeness,consistency and accuracy.

Conduct an exploratory factor analysis(EFA)with verimax rotation to determinethe underlying dimensions/constructs.

Test the constructs(identified in theprevious step) for a good fit to the data

using confirmatory factor analysis (CFA).

Develop and compare to each otherpredictive models to identify the best

performing model.

Conduct sensitivity analysis on the bestperforming predictive models to identify

the most important factors.

Step1

Step2

Step3

Step4

Step5

Step6

Fig. 2. Steps of the study methodology.

1153D. Delen et al. / Decision Support Systems 54 (2013) 1150–1160

In SVM, “attribute” is represented by predicted variable; “feature”is defined by transformed attribute. Hyperplane is defined by thetransformed attribute. Also “feature selection” is known as the taskof selecting the most appropriate representation. A set of featuresthat describe one case is called a “vector”.

SVM works by mapping data to a high dimensional feature space.The mapping functions can be either classification or regression func-tion. SVM belongs to the type of maximal margin classifier. There arefour kernel functions (linear function, polynomial function, radialbased function, and sigmoid function) to be used in classification prob-lems when the input data is not easily separable. To make the input dataeasily separable compared to the original data, the kernel functions areused to transform the input data to high dimensional feature space.

The aim of SVM is to find the optimal hyperplane that separatesthe clusters of vector in such a way that cases with one category ofthe target variable on one side of the plane and cases with the othercategory are on the other side of the plane. The vectors near thehyperplane are the support vectors. A separator which is drawn as ahyperplane is found between the separated classes. The ultimateaim of SVM is to establish a maximal margin between the separatedclasses. This will be able to offer a good classification performanceon the training data, and also provide high predictive accuracy forthe future data from the same distribution. The characteristic ofnew data after separation can be used for prediction. Since SVM'slearning ability is independent of the dimensionality of the futurespace, therefore SVM provides good performance [9].

3.3. Decision tree (C5 algorithm)

Decision trees are commonly used methods in data mining. Thetwo main types of classification generated by decision trees are clas-sification tree analysis and regression tree analysis. Decision treesare becoming increasingly popular methods of using data miningtechnique because decision trees are simple to understand and inter-pret, require little data preparation, handle numerical and categoricaldata and perform very well with large data set in a short time. Deci-sion trees demonstrate excellent visualization of results and relation-ships [3]. Although there are many specific decision tree algorithms,the ID3, C4.5, C5.0, CART, and CHAID (Chi-squared Automatic Interac-tion Detector) algorithms, based on the recent data mining literature,are arguably the most popular ones.

C5.0 is one of the most popular decision tree algorithms. Developedby Quinlan, C5.0 offers a number of improvements on its predecessorsC4.5 and ID3. It is significantly faster than ID3, and is more memory ef-ficient than C4.5. It creates considerably smaller decision tree while get-ting similar results to C4.5. It also supports boosting, which improves thetrees and gives themmore accuracy. C5.0 allows the weighting of differ-ent attributes and misclassification types. In addition, it automaticallyseparates the data to help reduce noise. Boosting is part of the C5.0 de-cision tree algorithm as an integration technology, which is included toimprove the accuracy of classification. C5.0 uses pre-pruning andpost-pruning methods to establish the decision tree.

C5.0 is a commercial and closed-source product [11,42]. It starts fromthe top level of the details to establish the decision tree. The set of train-ing examples is partitioned into two or more subsets based on the out-come of a test in the value of a single attribute. The particular test ischosen by an information theoretic heuristic that generally gives closeto optimal partitioning. This is repeated on each of the new subsetsuntil a subset contains only examples of a single class, or the partitioningtree has reached a predetermined maximum depth.

3.4. Logistic regression

Logistic regression is a generalization of linear regression. Regres-sion analysis is a statistical tool for the investigation of relationships.Like the linear regression analysis, most of the usual statistical

methods assume that the residuals, or errors, must follow a normaldistribution. If they are not the methods should not be used. Unlikeordinary linear regression, logistic regression does not assume thatthe dependent variable or the error terms are distributed normally.Also, it doesn't assume that the relationship between the indepen-dent variables and the dependent variable is linear. Logistic regres-sion is a variation of ordinary regression which is used when thedependent variable is a categorical variable.

Logistic regression also produces Odds Ratios (O.R.) associatedwith each predictor value. The “odds” of an event is defined as theprobability of the outcome event occurring divided by the probabilityof the event not occurring. The Odds Ratio for a predictor is defined asthe relative amount by which the odds of the outcome increase (O.R.greater than 1.0) or decrease (O.R. less than 1.0) when the value ofthe predictor variable is increased by 1.0 units.

4. Proposed methodology and its application

The individual steps of the methodology used in this study areshown on Fig. 2. Principal purposes of this methodology are: (1) to

1154 D. Delen et al. / Decision Support Systems 54 (2013) 1150–1160

find the most appropriate method for accurately predicting the finan-cial and non-financial performance; and (2) to identify the most im-portant knowledge management criteria (by using the developedprediction models as the source of the relationship between inputsand outputs) that improve the financial and non-financial perfor-mance; and by using these prioritized factors, propose recommenda-tions for the top management of the company in the service sector.

Step 1: The survey instrument regarding knowledge managementcriteria and organizational performance was developed.Step 2: In the second stage, the sample is selected and data is col-lected, examined and pre-processed.Step 3: An exploratory factor analysis (EFA) with varimax rotationwas employed to determine the underlying dimensions of knowl-edge management performance.Step 4: Knowledge management constructs were tested usingconfirmatory factor analysis (CFA) in order to determine if theextracted dimensions in step 4 offered a good fit to the data.Step 5: In this stage, the most popular classification methods suchas artificial neural networks, decision trees, support vector ma-chines and logistic regression techniques, are evaluated and thebest performing model specifications are selected.Step 6: After selecting the best model based on their predictive accu-racy mentioned above the most important knowledge managementvariables affecting the financial and non-financial performance in theservice sector were determined.

What follows is a number of sub-sections delineating the imple-mentation of these six steps.

4.1. Step 1: development of the survey instrument

Data were gathered via cross-sectional mail survey using aself-administered questionnaire that was essentially composed ofquestions related to KM practices and organizational performance.Respondents were asked to indicate the level of agreement basedon five-point Likert scales ranging from 1 “strongly disagree” to 5“strongly agree” on each of the items measuring eight aspects of KMpractices, which include organizational structure (OS), technologicalinfrastructure (TI), organizational culture (OC), intellectual capital(IC), knowledge generation (KG), knowledge codification and storage(KCS), knowledge transfer and sharing (KTS), and knowledge utilization(KU).

Measures of organizational non-financial and financial perfor-mance were based on items derived from a number of previous stud-ies using this variable [8,13,55]. The level of organizationalperformance measures was identified using judgmental measuresbased on managers' perceptions of how the organization performedon multiple indicators of organizational performance relative to its ri-vals based on a five-point scale, ranging from ‘muchworse than rivals’through ‘much better than rivals’. The non-financial performance in-dicators include: service quality as perceived by customer, marketshare gain over the last three years, reputation among major custom-er segments, capacity to develop a unique competitive profile, newproduct/service development, and market development. The financialperformance constructs include the following indicators; revenuegrowth over the last three years, net profits, return to investment,profit to revenue ratio, and cash flow from operation.

The original version of the questionnaire was in English. Thisquestionnaire was translated into the local language (Turkish). Thelocal version was back translated until a panel of experts agreedthat the two versions were comparable. The questionnaire waspre-tested several times to ensure that the wording, format, and se-quencing of questions were appropriate. As the percentage of missing

data was calculated to be relatively small, occasional missing data onvariables was handled by replacing them with the mean value.

4.2. Step 2: data collection

Privately-held service companies within the city of Istanbul inTurkey were selected as the sampling frame of this survey to investi-gate the most important knowledge management practices and tomeasure its effect on financial and non-financial performance. Forcenturies as being the largest city of Turkey, Istanbul has nearlyone-fifth of the nationwide population and has been undisputedlythe main industrial and trade center. Data for this study was collectedbased on a self-administered questionnaire that was distributed tothe Chief Administrative Officers of privately-held service companiesthat includes finance, logistic, stationary sectors, within the city ofIstanbul. Of the 1200 questionnaires posted, a total of 738 usablequestionnaires were returned after one follow-up, comprising aresponse rate of 61.5%. The responses indicated that a majority ofthe respondents completing the questionnaire were in fact membersof the top management. It was requested that the questionnaire becompleted by a senior officer/executive in charge of HRM and KMpractices. A test for non-response bias for the mail survey was alsoconducted by comparing the first wave of survey responses to thelast wave of survey responses. The test results indicated no significantdifference in the responses between early and late respondents(p>0.1). Therefore, no response bias was evident.

4.3. Step 3: exploratory factor analysis (EFA)

Exploratory factor analysis with varimax rotation was performedon the Knowledge Management Infrastructure criteria in order toextract the dimensions underlying the construct. The EFA of the 29variables yielded five factors explaining 62.8% of the total variance.Based on the items loading on each factor, these factors were labeledas ‘technology’ (Factor 1), ‘leadership’ (Factor 2), ‘human capital’(Factor 3), ‘organizational culture’ (Factor 4), and ‘organization struc-ture’ (Factor 5). These items are shown in Table 1. The Cronbach alphavalues of reliability for the underlying factors range from 0.72 to 0.92suggesting satisfactory level of construct reliability [40].

Similarly, EFA was undertaken to produce a set of parsimoniousdistinct non-overlapping dimensions of Knowledge ManagementProcesses from the full set of 33 items. The factor analysis produced5 factors which explained 66.4% of the observed variance, as shownin Table 2. These factors were labeled as knowledge utilization (Factor1), knowledge classification and coding (Factor 2), knowledge gener-ation (Factor 3), formal knowledge sharing (Factor 4) and informalknowledge sharing (Factor 5). The Cronbach alpha values of reliabili-ty for the underlying factors range from 0.73 to 0.93 suggestingsatisfactory level of construct reliability.

4.4. Step 4: confirmatory factor analysis (CFA)

This stage is also known as testing the measurement model, wherethe constructs of Knowledge Management Infrastructure and Knowl-edge Management process are tested using the first order confirmato-ry factor model to assess construct validity using the method ofmaximum likelihood. The results consistently supported the factorstructure for two constructs as discussed earlier in the EFA stage.The confirmatory factor analysis (CFA) technique is based on thecomparison of variance-covariance matrix obtained from the sampleto the one obtained from the model. The measurement model resultsare presented in Tables 3 and 4, respectively.

The figures in Tables 3 and 4 exhibit the standardized regressionweight between each manifest variable and its corresponding latentvariable. It was found that all t-values in the CFA are statistically

Table 1EFA of the KM infrastructure.

Variables Factors

1 2 3 4 5

Information systems in our corporate are convenient for our needs. 0.82Information technology systems in our corporate are new and fast. 0.80Our corporation has an efficient information system to be used for KM strategies. 0.78Our corporation has adequate database systems. 0.75Our corporate adequately invests in IT. 0.75The internet and intranet have been effectively used in our corporate. 0.74KM systems in our corporate (e.g. software regarding information saving and retrieval, databases, and search engines) are user-friendly. 0.67Emailing systems have been effectively used in our corporate. 0.59Our managers encourage us to learn more about KM. 0.76Our managers are good representatives of KM implementers. 0.74Our managers are supportive in developing, using, and sharing the knowledge. 0.73Our managers support creating new ideas and brainstorming. 0.73Our managers support us in our knowledge management-related activities. 0.73Our managers possess the required experience and competency in terms of knowledge management (KM). 0.72In our corporation, top management weighs much importance on the KM. 0.63Managers and employees of our corporate are experienced in their jobs. 0.78Managers and employees of our corporate have enough technical knowledge in their domains. 0.78Employees of our corporate both at the personnel and management level possess the competence in terms of knowledge and personal features. 0.75Employees of our corporate are qualified at a satisfactory level in terms of their educational background. 0.72I can identify employees of our corporate as “highly qualified knowledge management individuals”. 0.66Our corporate conducts adequate number of training activities. 0.44Our corporate culture encourages teamwork. 0.76Our corporate culture supports the idea of cooperation and knowledge sharing. 0.74Our corporate culture encourages knowledge creation. 0.62We trust in our colleagues and managers. 0.53There exists an effective delegation of authority and the managers of each unit can make their own departmental decisions without consultingthe top management.

0.72

Without any hesitation, I can do my own initiative regarding my job within my authority. 0.71There is not a rigid chain of command between different levels of management. 0.61There are no problems in terms of establishing and sharing authority and responsibility. 0.46

Table 2EFA of the KM process.

Variables Factors

1 2 3 4 5

Corporate knowledge is reflected in customer relationship processes. 0.79Corporate knowledge is reflected in our production and service systems. 0.78Our corporate has a management style which is convenient to practically make use of the accumulated knowledge. 0.73Our decision making processes are working efficiently. 0.70Our corporate has adopted the philosophy of “continuous learning” and practicing the lessons learned. 0.69The knowledge obtained through the training is implemented in a short while. 0.68Our corporate effectively makes use of its knowledge potential. 0.67I effectively make use of my knowledge/experience in the work environment. 0.66We have data storage and archiving system through which we can get rapid access to accurate information. 0.80All personnel of our corporate record the data and the information which is revealed by their operations. 0.76We have an effective record-storing system in which the related information about the products, services, employees, and customers is saved. 0.75Information about the products, services, employees, and customers is updated on a regular basis. 0.75Our duties and operations are well-identified and are recorded. 0.75All information about my job is regularly classified, filed, and stored in an electronic environment. 0.71I can quickly and easily obtain the information that I need. 0.69The information about our suppliers and competitors is saved and recorded up-to-date. 0.63Our corporate supports innovative thinking and encourages creating innovative ideas. 0.75All employees of our corporate are encouraged for continuous learning. 0.73Highly qualified and featured individuals are tried to be attracted to work in our corporate. 0.72Our corporate successfully implements a suggestion system. 0.66Brainstorming sessions are conducted for creating alternative solutions to problems and for system development (i.e. improving the current systems). 0.65Employees of our corporate actively contribute to the process of knowledge creation. 0.64Our corporate conducts enough number of R&D activities. 0.63I have access to new and updated information on the web. 0.61Our corporate is systematically devoted to create and develop knowledge. 0.60We pay special attention to share the accumulated knowledge with our colleagues. 0.77We improve our work processes by sharing our knowledge and experience with our colleagues. 0.75Teamwork is very helpful in sharing the knowledge. 0.65We efficiently make use of e-mailing and the Internet to share the knowledge. 0.51To achieve informal knowledge-sharing, we organize picnics, soccer games, and family visits with our friends and try to get together apart fromthe work environment.

0.80

We conduct meetings with the other departments to coordinate the knowledge-sharing. 0.60There is a strong communication between the employees and the managers. 0.55

1155D. Delen et al. / Decision Support Systems 54 (2013) 1150–1160

Table 3CFA of the KM infrastructure.

Variables Regression weight t-Value

TechnologyInformation systems in our corporate are convenient for our needs. 0.89⁎ 28.1Information technology systems in our corporate are new and fast. 0.81⁎ 24.7Our corporation has an efficient information system to be used for KM strategies. 0.73⁎ 21.7Our corporation has adequate database systems. 0.68⁎ 19.6Our corporate adequately invests in IT. 0.80⁎ –

The internet and intranet have been effectively used in our corporate. 0.64⁎ 18.5KM systems in our corporate (e.g. software regarding information saving and retrieval, databases,and search engines) are user-friendly.

0.73⁎ 21.5

Emailing systems have been effectively used in our corporate. 0.81⁎ 16.3

LeadershipOur managers encourage us to learn more about KM. 0.77⁎ 21.9Our managers are good representatives of KM implementers. 0.76⁎ –

Our managers are supportive in developing, using, and sharing the knowledge. 0.79⁎ 22.3Our managers support creating new ideas and brainstorming. 0.72⁎ 20.1Our managers support us in our knowledge management-related activities. 0.72⁎ 20.2Our managers possess the required experience and competency in terms of knowledge management (KM). 0.73⁎ 20.7In our corporation, top management weighs much importance on the KM. 0.67⁎ 18.5

Human capitalManagers and employees of our corporate are experienced in their jobs. 0.78⁎ 16.7Managers and employees of our corporate have enough technical knowledge in their domains. 0.79⁎ 16.9Employees of our corporate both at the personnel and management level possess the competence in terms ofknowledge and personal features.

0.81⁎ 17.4

Employees of our corporate are qualified at a satisfactory level in terms of their educational background. 0.66⁎ 16.9I can identify employees of our corporate as “highly qualified knowledge management individuals”. 0.63⁎ –

Our corporate conducts adequate number of training activities. 0.43⁎ 10.5

Organizational cultureOur corporate culture encourages teamwork. 0.79⁎ 24.9Our corporate culture supports the idea of cooperation and knowledge sharing. 0.75⁎ –

Our corporate culture encourages knowledge creation. 0.78⁎ 19.9We trust in our colleagues and managers. 0.63⁎ 16.2

Organizational structureThere exists an effective delegation of authority and the managers of each unit can make their owndepartmental decisions without consulting the top management.

0.44⁎ 10.7

Without any hesitation, I can do my own initiative regarding my job within my authority. 0.57⁎ 13.8There is not a rigid chain of command between different levels of management. 0.70⁎ 16.6There are no problems in terms of establishing and sharing authority and responsibility. 0.73⁎ –

– Fixed for estimation.⁎ pb0.01.

1156 D. Delen et al. / Decision Support Systems 54 (2013) 1150–1160

significant at 0.01 levels. It indicates that all the individual factor load-ings to be highly significant, giving support to convergent validity.

The goodness-of-fit indices for Knowledge Management Infra-structure and Knowledge Management process are demonstrated inTable 5. These indices conform to the normal acceptable standards.The values of χ2 statistic were 1133 and 1615, with the values ofχ2/df ratio varying between 3.16 and 3.52 for the KnowledgeManagement Infrastructure and Knowledge Management processconstructs respectively. This ratio should be within the range of 0–5where lower values indicate a better fit. The results show that con-structs do fit in with this criterion. In addition, the GFI, AGFI and CFIfor the Knowledge Management Infrastructure and Knowledge Man-agement process constructs are highly satisfactory, as they are veryclose to a value of 1.0, which denotes a perfect fit. The results attestthe construct validity for the measurement models of KnowledgeManagement Infrastructure and Knowledge Management process.

4.5. Step 5: comparative analysis of predictive models

In the study, total ten inputs and two outputs were employed. Theseten inputs are ‘technology’, ‘leadership’, ‘human capital’, ‘organizationalculture’, ‘organization structure’, ‘knowledge utilization’, ‘knowledgeclassification and coding’, ‘knowledge generation’, ‘formal knowledgesharing’ and ‘informal knowledge sharing’. Two outputs, ‘financial

performance’ and ‘non-financial performance’, are selected as binaryvariables.

Average values for the financial and non-financial organizationperformance were used as a split criterion. The group with a perfor-mance score of above average value was rated as 1 and the groupwith the performance score of below average value was rated as 0.With this information, individual organization can be classified assuccessful and not successful.

The performance of the models used in binary (two-group) ismeasured by using a confusion matrix which is given in Table 6. Aconfusion matrix contains valuable information about actual and pre-dicted classifications created by the classification model. For the pur-poses of the study, we used well-known performance measures suchas overall accuracy, AUC (Area Under the ROC Curve), Recall andF-measure. All of these measures for each model used in the studywere evaluated, after which the models were compared on the basisof proposed performance measurements.

For the SVMmodel, the linear, polynomial, sigmoid and RBF (Radi-al Basis Functions) kernel functions were tested. For the financial andnon-financial performance models, SVM Linear function and SVM(RBF) function were selected respectively.

The data set was partitioned into training and testing data sets.70% of the data was used for training and 30% was used fortesting. For performance analysis, the test data sets were used forassessment.

Table 4CFA of the KM process.

Variables Regression weight t-Value

Knowledge utilizationCorporate knowledge is reflected in customer relationship processes. 0.75⁎ 25.3Corporate knowledge is reflected in our production and service systems. 0.85⁎ 26Our corporate has a management style which is convenient to practically make use of the accumulated knowledge. 0.85⁎ 26.2Our decision making processes are working efficiently. 0.80⁎ 24.4Our corporate has adopted the philosophy of “continuous learning” and practicing the lessons learned. 0.78⁎ –

The knowledge obtained through the training is implemented in a short while. 0.78⁎ 23.4Our corporate effectively makes use of its knowledge potential. 0.81⁎ 24.6I effectively make use of my knowledge and experience in the work environment. 0.72⁎ 21.2

Knowledge classification and codingWe have data storage and archiving system through which we can get rapid access to accurate information. 0.79⁎ 23All personnel of our corporate record the data and the information which is revealed by their operations. 0.81⁎ 23.7We have an effective record-storing system in which the related information about the products, services, employees,and customers is saved.

0.77⁎ 22.2

Information about the products, services, employees, and customers is updated on a regular basis. 0.82⁎ 24Our duties and operations are well-identified and are recorded. 0.79⁎ 22.9All information about my job is regularly classified, filed, and stored in an electronic environment. 0.77⁎ –

I can quickly and easily obtain the information that I need. 0.78⁎ 26.9The information about our suppliers and competitors is saved and recorded up-to-date. 0.69⁎ 19.6

Knowledge generationOur corporate supports innovative thinking and encourages creating innovative ideas. 0.80⁎ 22.3All employees of our corporate are encouraged for continuous learning. 0.73⁎ 20.2Highly qualified and featured individuals are tried to be attracted to work in our corporate. 0.72⁎ 19.5Our corporate successfully implements a suggestion system. 0.68⁎ 18.6Brainstorming sessions are conducted for creating alternative solutions to problems and for system development(i.e. improving the current service and production systems).

0.74⁎ 24.7

Employees of our corporate actively contribute to the process of knowledge creation. 0.75⁎ –

Our corporate conducts enough number of R&D activities. 0.56⁎ 15I have access to new and updated information on the web. 0.70⁎ 19.2Our corporate is systematically devoted to create and develop knowledge. 0.77⁎ 25.1

Formal knowledge sharingWe pay special attention to share the accumulated knowledge with our colleagues. 0.77⁎ –

We improve our work processes by sharing our knowledge and experience with our colleagues. 0.73⁎ 26.8Teamwork is very helpful in sharing the knowledge. 0.82⁎ 21.8We efficiently make use of e-mailing and the Internet to share the knowledge. 0.61⁎ 16.1

Informal knowledge sharingTo achieve informal knowledge-sharing, we organize picnics, soccer games, and family visits with our friends and try toget together apart from the work environment.

0.53⁎ 13.3

We conduct meetings with the other departments to coordinate the knowledge-sharing. 0.72⁎ 19.6There is a strong communication between the employees and the managers. 0.84⁎ –

– Fixed for estimation.⁎ pb0.01.

1157D. Delen et al. / Decision Support Systems 54 (2013) 1150–1160

The Area Under the Curve (AUC) for the test data sets was used tocompare the mode's predictive performances. According to the AUC,the SVM–RBF and SVM–Linear models demonstrated very good per-formance measurements, with the C5.0 decision tree being the bestof all the presented models.

4.5.1. Overall accuracy (AC)Accuracy is known as the percentage of records that is correctly

predicted by the model. It is defined as being the ratio of correctlypredicted cases to the total number of cases.

Accuracy ¼ TPþ TNTPþ TNþ FPþ FN

ð1Þ

Table 5Goodness-of-fit statistics.

Model/construct χ2 χ2/df RMR GFI AGFI CFI

KM process 1133 3.16 0.06 0.91 0.88 0.94KM infrastructure 1625 3.62 0.055 0.89 0.86 0.93

4.5.2. PrecisionPrecision is defined as the ratio of the number of True Positive (cor-

rectly predicted cases) to the sumof the True Positive and False Positive.

4.5.3. RecallRecall is also known as the Sensitivity or True Positive rate. It is de-

fined as the ratio of the True Positive (the number of correctly pre-dicted cases) to the sum of the True Positive and the False Negative.

4.5.4. F-measureF-measures take the harmonic mean of the Precision and Recall

Performance measures. Therefore, it takes into consideration boththe Precision and the Recall as being important measurement toolsfor these calculations [3].

F‐measure ¼ 2� Precision� RecallPrecisionþ Recall

ð2Þ

According to the overall accuracy rate, the SVM–Linear model out-performs three models in accuracy (AC), but not significantly. As a re-sult of AC evaluation, it achieves a higher accuracy rate of 77.8%.SVM–RBF model accuracy rate is very close to SVM–Linear but it is

Table 6Confusion matrix for financial performance.

Predicted

Actual

Not Successful Successful

Not Successful True Negative False Positive

Successful False Negative True Positive

1158 D. Delen et al. / Decision Support Systems 54 (2013) 1150–1160

not so significant. Following these models, the next best model thatdemonstrates a high accuracy rate is the C5.0 decision tree algorithm(77.4%).

According to the overall accuracy rate, SVM–Linear outperformsthree models in accuracy (AC) even if it is not significantly. SVM–

Linear achieves the highest accuracy rate (77.8%). SVM–RBF accuracyrate is very close to SVM–Linear in terms of performance. In addition,the next best model that demonstrates high accuracy rate is the C5.0decision tree algorithm (77.4%).

In terms of precision (P), SVM–Linear model does outperform theother models significantly. SVM Linear achieves 80.2% precision ratewhich is significantly higher than the rest of the compared models.When the F-measure is considered, the C5.0 decision tree model out-performs the rest of the models (79.1%), while the SVM–Linear modelachieves the second best F-measure rate (76.6%). Table 7 comparesfive models in considering test data set. The results indicated thatSVM–Linear model outperformed the other models.

On the other hand, considering the non-financial performancemeasurement, SVM (RBF) model performs best in terms of overall ac-curacy rate (82.4%) which is shown in Table 8. It outperforms all theavailable models. In addition to accuracy rate, SVM model has thehighest area under curve (.88). When we look at recall performancemeasurement, SVM and logistic regression perform equally well.They both possess 82.3% in recall. Furthermore, SVM model outper-forms three models in F-measure as well. It achieves the highestF-measure with 83.5%. Even if neural network performs the highestprecision measurement with 84.9%, it is not significantly higherthan SVM model's precision of 84.6%. In summary, SVM (RBF)model performs very well for non-financial performance.

4.6. Step 6: determining the most important KM variables

As mentioned in the previous steps, SVM–Linear model has anacceptable capability of predicting the organizational financial

Table 7Prediction results for financial performance.

Accuracy (AC) Sensitivity/True Positiverate or recall (TP)

False Positirate (FP)

SVM (Linear) 0.778 0.733 0.178Decision tree (C5.0) 0.774 0.867 0.318Logistic regression 0.764 0.724 0.196Neural network 0.755 0.762 0.252

Table 8Prediction results for non-financial performance.

Area Under theCurve (AUC)

Accuracy (AC) Sensitivity/True PositiveRate or Recall (TP)

FR

SVM (RBF) 0.880 0.824 0.825 0C5.0 0.840 0.803 0.802 0Logistic regression 0.878 0.811 0.825 0Neural network 0.878 0.815 0.802 0

performance. The impact of KM dimensions on organizational finan-cial performance was evaluated and ranked based on the samemodel. This also provides managers with invaluable information inidentifying which KM practices they should concentrate in order tohave a better organizational financial performance. Table 9 showsthe contribution of KM practices to the organizational financial per-formance in terms of the degree of their importance levels and theirrespective rankings. From the full set of 10 KM practices, knowledgeutilization (0.34) appears as a leading factor. Organizational Cultureis the second most important criterion with the importance level of0.237, while knowledge generation is found to be the third importantKM practice with the importance level of 0.176. The fourth importantKM practices is informal knowledge sharing. In contrast, human capi-tal and organizational structure are the least important KM practices interms of their effects on organizational financial performance. Thisfinding is not particularly surprising that KM practices have been pri-marily focused on knowledge utilization and knowledge generation andinformal sharing in service industry.

Table 10 shows the most important KM dimensions on organiza-tional non-financial performance. Similarly with the financial perfor-mance, it was discovered that knowledge utilization has been foundto be the most important KM practices in terms of their effects onnon-financial organizational performance. Organizational structureappeared to be the second important factor, which is also consistentwith the existing KM literature. Similarly, knowledge generationwas found to be the third important critical factor affecting thenon-financial organization performance. Technological Infrastructurealso featured as important though they had relatively less impact onnon-financial performance.

5. Summary and conclusion

There is a scant research attention which considers the knowledgemanagement practices and non-financial performance and financialperformance within the context of service industry. The main thrustof this study was to investigate and determine the most importantknowledge management practices in terms of both financial andnon-financial performance using machine learning tools.

Exploratory and confirmatory factor analyses were employed toproduce empirically verified and validated underlying dimensions ofKM practice construct drawing on a sample of service sector. Thefindings of study indicated that knowledge utilization had a strongand positive effect on both non-financial and financial performance.Also a strong and positive relationship was noted between knowl-edge generation and non-financial performance and financial

ve Specificity/TrueNegative rate (TN)

False Negativerate (FN)

Precision (P) F measure

0.822 0.267 0.802 0.7660.682 0.133 0.728 0.7910.804 0.276 0.784 0.7520.748 0.238 0.748 0.755

alse Positiveate (FP)

Specificity/TrueNegative rate (TN)

False Negativerate (FN)

Precision (P) F measure

.178 0.822 0.175 0.846 0.835

.196 0.804 0.198 0.828 0.815

.206 0.794 0.175 0.825 0.825

.168 0.832 0.198 0.849 0.824

Table 9Importance of KM practices on the organizational financial performance.

KM practices Importance level Ranking

Knowledge utilization (KU) 0.34 1Organizational culture (OC) 0.237 2Knowledge generation (KG) 0.176 3Informal knowledge sharing 0.106 4Human capital 0.089 5Organizational structure (OS) 0.004 6

1159D. Delen et al. / Decision Support Systems 54 (2013) 1150–1160

performance. Finally, the results provided empirical evidence that theorganizational culture and organizational structure were found as im-portant for financial performance and non-financial performancerespectively.

Third generation KM studies particularly concentrates on the pos-sible effects of KM applications on organizational performance. Orga-nizational performance on the other hand, has been analyzed fromtwo stand points which are financial performance and non-financialperformance. Many research findings suggest that the relationshipbetween KM and non-financial organizational performance is moresignificant and direct than the relationship between KM and financialperformance [18,54].

Our study also analyzed the effects of KM on financial andnon-financial organizational performance. The research findings re-veal that knowledge utilization appears to be the most important fac-tor of KM in terms of its effects on both financial and non-financialperformance. It is mainly because knowledge is valuable if only it isutilized. Hence, one of the most challenging dimensions of KM ishow to leverage and utilize knowledge in accordance with organiza-tional objectives and convert it into a valuable form in order to gaincompetitive advantage. In our research we also found that the effectsof knowledge utilization are more significant in non-financial perfor-mance than financial performance which is also convenient with thefindings.

When we look at the effects of KM on organizations' financial per-formance, knowledge utilization and organizational culture are oftenfound to be the most significant factors of KM followed by knowledgegeneration, informal knowledge sharing, and human capital. Organi-zational structure has ignorable effect on financial performance.

As mentioned before knowledge utilization is also the most im-portant factor of KM from the non-financial stand point as well.Furthermore, other important factors affecting the non-financialperformance are organizational structure, knowledge generation,technological infrastructure, and knowledge classification and coding.The importance of organizational culture, informal knowledgesharing, and leadership is less in comparison to previous factors.

As is the case in any research, there are a number of limitations ofthis study. The most important limitation of this study is that it com-prises only one sector in a specific geographic area in Turkey. Further-more, even though the sample size seems to be satisfactory, a largernumber of participants would have made the study stronger. Thefuture directions of this research will focus on mitigating these

Table 10Importance of KM practices on the organizational non-financial performance.

KM practices Importance level Ranking

Knowledge utilization 0.477 1Organizational structure 0.133 2Knowledge generation 0.122 3Technological infrastructure 0.099 4Knowledge classification and coding 0.062 5Organizational culture 0.04 6Informal knowledge sharing 0.039 7Leadership 0.028 8

limitations by sampling SMEs from other sectors and geographicregions to obtain a larger, more comprehensive sample data.

References

[1] R. Beijerse, Questions in knowledge management: defining and conceptualizing aphenomenon, Journal of Knowledge Management 3 (2) (1999) 94–110.

[2] S. Bender, A. Fish, The transfer of knowledge and the retention of expertise: thecontinuing need for global assignments, Journal of Knowledge Management 4(2) (2000) 125–135.

[3] J.A.M. Berry, G. Linoff, Data Mining Techniques—For Marketing, Sales and Custom-er Support, Wiley Publishing, New York, 1997.

[4] G. Bhatt, Organizing knowledge in the knowledge development cycle, Journal ofKnowledge Management 4 (1) (2000) 15–26.

[5] G. Bhatt, Knowledge management in organizations: examining the interactionsbetween technologies, techniques, and people, Journal of Knowledge Manage-ment 5 (1) (2001) 68–75.

[6] M. Boisot, Knowledge Assets, Oxford University Press, New York, 1998.[7] N. Bontis, W. Keow, S. Richardson, Intellectual capital and business performance

in Malaysian industries, Journal of Intellectual Capital 1 (1) (2000) 85–100.[8] K.B. Boyd, Strategic planning and financial performance: a meta-analytic review,

Journal of Management Studies 28 (1991) 353–374.[9] N. Cristianini, J. Shawe-Taylor, An Introduction to Support Vector Machines and

Other Kernel-based Learning Methods, Cambridge University Press, London,2000.

[10] T. Davenport, L. Prusak, Working Knowledge, Harvard Business School, Boston,MA, 1998.

[11] D. Delen, A comparative analysis of machine learning techniques for studentretention management, Decision Support Systems 49 (2010) 498–506.

[12] M. Demirbag, S.C.L. Koh, E. Tatoglu, S. Zaim, TQM and market orientation's impacton SMEs' performance, Industrial Management & Data Systems 106 (8) (2006)1206–1228.

[13] G.G. Dess, R.B. Robinson Jr., Measuring organizational performance in the absenceof objective measures: the case of the privately-held firm and conglomeratebusiness unit, Strategic Management Journal 5 (1984) 265–273.

[14] C. Drott, Personal knowledge, corporate information: the challenges for compet-itive intelligence, Business Horizons 44 (2) (2001) 31–37.

[15] P.M. Fayard, Strategic communities for knowledge creation: a Western proposalfor the Japanese concept of Ba, Journal of Knowledge Management 7 (5) (2003)25–31.

[16] S. Firer, M. Williams, Intellectual capital and traditional measures of corporateperformance, Journal of Intellectual Capital 4 (3) (2003) 348–360.

[17] E. Geisler, A typology of knowledge management: strategic groups and role be-havior in organizations, Journal of Knowledge Management 11 (1) (2007) 84–96.

[18] I. Goll, N.B. Johnson, A.A. Rasheed, Knowledge capability, strategic change, andfirm performance: the moderating role of the environment, Management Deci-sion 45 (2007) 161–179.

[19] J. Guthrie, U. Johanson, P.N. Bukh, P. Sanchez, Intangibles and the transparententerprise: new strands of knowledge, Journal of Intellectual Capital 4 (4) (2003)429–440.

[20] S. Harrison, P. Sullivan, Profiting from intellectual capital, Journal of IntellectualCapital 1 (1) (2000) 33–46.

[21] S. Hauschild, T. Licht, W. Stein, “Creating a Knowledge Culture”, McKinseyQuarterly. accessed at www.mckinseyquarterly.com 2002.

[22] S.O. Haykin, Neural Networks and Learning Machines, Prentice Hall, New Jersey,2009.

[23] T.H.O. Herrgard, Difficulties in diffusion of tacit knowledge in organizations,Journal of Intellectual Capital 1 (4) (2000) 357–365.

[24] R. Hicks, R. Dattero, S. Galup, A metaphor for knowledge management: explicitislands in a tacit sea, Journal of Knowledge Management 11 (1) (2007) 5–16.

[25] J. Hong, Structuring for organizational learning, The Learning Organization 6 (4)(1999) 173–185.

[26] M. Jacop, G. Ebrahimpur, Experience vs. expertise, Journal of Intellectual Capital 2(1) (2001) 74–88.

[27] T. Kalling, Knowledge management and the occasional links with performance,Journal of Knowledge Management 7 (3) (2003) 67–81.

[28] T. Koulopoulos, C. Frappaolo, Smart Things to Know About Knowledge Manage-ment, Capstone Publishing Limited, United Kingdom, 1999.

[29] V.G. Krogh, J. Roos, Managing Knowledge, Sage, London, UK, 1996.[30] J.C. Lang, Managerial concerns in knowledge management, Journal of Knowledge

Management 5 (1) (2001) 43–59.[31] S. Lee, Using data envelopment analysis and decision trees for efficiency analysis

and recommendation of B2C controls, Decision Support Systems 49 (2010)486–497.

[32] C. Lueg, Information, knowledge and networked minds, Journal of KnowledgeManagement 5 (2) (2001) 151–159.

[33] M. Lytras, A. Pouloudi, A. Poulymenakou, Knowledge management convergence —expanding learning frontiers, Journal of Knowledge Management 6 (1) (2002)40–51.

[34] R. Maier, T. Hadrich, Centralized versus peer-to-peer knowledge managementsystems, Knowledge and Process Management 13 (1) (2006) 47–61.

[35] Y. Malhotra, Knowledge management for E-business performance: advancinginformation strategy to Internet time, in: Information Strategy: The Executives'Journal, Summer issue, 2000, pp. 8–16.

1160 D. Delen et al. / Decision Support S

[36] Y. Malhotra, Integrating knowledge management technologies in organizationalbusiness processes: getting real time enterprises to deliver real business perfor-mance, Journal of Knowledge Management 9 (1) (2005) 7–28.

[37] B. Marr, O. Gupta, S. Pike, G. Roos, Intellectual capital and knowledge managementeffectiveness, Management Decision 41 (8) (2003) 771–781.

[38] R. McAdam, S. Mcreedy, A critical review of knowledge management models, TheLearning Organization 6 (3) (1999) 91–100.

[39] H. Nemati, Global knowledge management: exploring a framework for research,Journal of Global Information Technology Management 5 (3) (2002) 1–11.

[40] J.C. Nunnally, Psychometric Theory, 2nd ed. McGraw-Hill, New York, 1978.[41] C. O'Dell, J. Grayson, If Only We Knew What We Know: The Transfer of Internal

Knowledge and Best Practice, The Free Press, U.S.A., 1998.[42] D.L. Olson, D. Delen, Y. Meng, Comparative analysis of data mining methods for

bankruptcy prediction, Decision Support Systems 52 (2012) 464–473.[43] C. Ordaz, M. Allez, F. Alcazar, P.M. Fernandez, R. Cabrera, Internal diversification

strategies and the process of knowledge creation, Journal of KnowledgeManagement 8 (1) (2004) 77–93.

[44] S. Robertson, A tale of knowledge-sharing systems, Journal of KnowledgeManagement 6 (3) (2002) 295–308.

[45] D. Skyrme, Knowledge Networking, Butterworth-Heinemann, Oxford, 1999.[46] D. Skyrme, Measuring Intellectual Capital. accessed at www.skyrme.com2002.[47] T. Stewart, The Wealth of Knowledge: Intellectual Capital and the Twenty-First

Century Organization, Doubleday, Random House Inc., U.S.A., 2001.[48] K.E. Sveiby, R. Simons, Collaborative climate and effectiveness of knowledge work—

an empirical study, Journal of Knowledge Management 6 (5) (2002) 420–433.[49] K. Toften, S. Olsen, Export market information use, organizational knowledge and

firm performance, International Marketing Review 20 (1) (2003) 95–110.[50] UNCTAD, Small- and Medium-sized Transnational Corporations, United Nations,

New York, 1993.[51] V. Vapnik, The Nature of Statistical Learning Theory, Springer-Verlag, New York,

1995.[52] P. Wilhelmij, R. Schmidt, Where does knowledge management add value, Journal

of Intellectual Capital 1 (4) (2000) 366–380.[53] B. Yilmaz, Küçük ve orta büyüklükteki işletmelerin toplumda üstlendikleri roller

bakımından analizi, Dış Ticaret Dergisi 9 (30) (2004) 141–179.[54] M. Zack, J. McKeen, S. Singh, Knowledge management and organizational

performance: an exploratory analysis, Journal of Knowledge Management 13 (6)(2009) 392–409.

[55] H. Zaim, Knowledge management implementation in IZGAZ, Journal of Economicand Social Research 8 (2006) 1–25.

[56] H. Zaim, E. Tatoglu, S. Zaim, Performance of knowledge management practices: acausal analysis, Journal of Knowledge Management 11 (6) (2007) 54–67.

[57] X. Zhao, J. Xie, The impact of information sharing and ordering co-ordination onsupply chain performance, Supply Chain Management 7 (1) (2002) 24–40.

Dr. Dursun Delen is theWilliam S. Spears Chair in BusinessAdministration and Professor of Management Science andInformation Systems in the Spears School of Business atOklahoma State University (OSU). He received his Ph.D. inIndustrial Engineering and Management from OSU in1997. Prior to his appointment as an Assistant Professorat OSU in 2001, he worked for a privately-owned researchand consultancy company, Knowledge Based Systems Inc.,in College Station, Texas, as a research scientist for fiveyears, during which he led a number of decision supportand other information systems related research projectsfunded by federal agencies, including DoD, NASA, NISTand DOE. His research has appeared in major journals

including Decision Support Systems, Decision Sciences,

Communications of the ACM, Computers and Operations Research, Computers in Indus-try, Journal of Production Operations Management, Artificial Intelligence in Medicine,Expert Systems with Applications, among others. He recently published four books:Advanced Data Mining Techniques with Springer, 2008; Decision Support and BusinessIntelligence Systems with Prentice Hall, 2010; Business Intelligence: A ManagerialApproach, with Prentice Hall, 2010; and Practical Text Mining and Statistical Analysisfor Non-structured Text Data Applications, with Elsevier, 2012. He is often invited tonational and international conferences for keynote addresses on topics related toData/Text Mining, Business Intelligence, Decision Support Systems, and KnowledgeManagement. He served as the general co-chair for the 4th International Conferenceon Network Computing and Advanced Information Management (September 2–4,2008 in Soul, South Korea), and regularly chairs tracks and mini-tracks at various infor-mation systems conferences. He is the associate editor-in-chief for International Journalof Experimental Algorithms, associate editor for International Journal of RF Technolo-gies, and is on editorial boards of five other technical journals. His research and teach-ing interests are in data and text mining, decision support systems, knowledgemanagement, business intelligence and enterprise modeling.

Dr. Halil Zaim obtained his Bachelor degree in Economicsfrom Istanbul University. He also completed his Masterand Ph.D. education in Labor Economics from the sameuniversity. He is an Associate Professor at Fatih University,Management Department and Director of ContinuousEducation Center of the University. Zaim has three pub-lished books, numerous national and international journalpapers and congress proceedings. His current scholarly in-terest is on Human Resource Management, KnowledgeManagement, and Business Ethics. Zaim is married withtwo children.

ystems 54 (2013) 1150–1160

Dr. Cemil Kuzey lectures at the Department of Manage-ment at Fatih University in Istanbul, Turkey, teachingOperation Research and Statistics for Social Sciences. He ac-quired his Ph.D. degree in Business Administration throughthe Department of Quantitative Analysis, Istanbul Universi-ty, Turkey. Among his academic pursuits, he took severalgraduate courses at the Ontario Institute for Studies inEducation, University of Toronto. His research interestsare related to Operation Research, Data Mining, PredictiveModeling and Business Intelligence.

Dr. Selim Zaim has received his B.S. degree in MechanicalEngineering from Istanbul Technical University and hisPh.D. degree in Production and Operations Managementfrom Istanbul University. He has been serving as a profes-sor in the Faculty of Technology at Marmara University.He has published over 100 articles and papers in variousjournals and congress proceedings. His current scholarlyinterests focus on multivariate data analysis, supply chainmanagement, data mining and multi-criteria decision mak-ing. He reviews papers for a variety of journals. He is amember of the Industrial Management and DevelopmentAssociations and Quality Association in Turkey (KALDER).