A hybrid approach for market segmentation and market segment evaluation and selection: An integration of Data mining and MADM Mohammad Hasan Aghdaie1,2 Sarfaraz Hashemkhani Zolfani1,2 Edmundas Kazimieras Zavadskas11Vilnius Gediminas Technical University, Institute of Internet and Intelligent Technologies Sauletekio al. 11, LT-10223 Vilnius, Lithuania2Shomal University, Department of Industrial Engineering, P. O. Box 731, Amol, Mazandaran, IranEmails:********************************************************************** AbstractDecision making in marketing becomes more and more sophisticated and two important issues in marketing decisions are market segmentation and segment evaluation and selection. These decisions are two focal points of all companies which many strategies are followed or influenced by them. Decision making is based on information and data mining aims to extract useful information form implicit, unknown and raw data. Clustering techniques are one of the widely used data mining tools which have been used for dividing market into different segments. In recent years, numerous papers about using data mining or multi attribute decision making (MADM) for marketing decisions have been published. MADM tools are used as a natural approach for evaluating alternatives with respect to conflict criterion. Thus, this study aims to integrate these two fields for improving quality of market segmentati on’s decisions. The proposed methodology is a combination of data mining tools for market segmentation and MADM tools for evaluation and selection of the best market. More precisely, clustering is used to divide a whole market into different segments. Two MADM tools including, step-wise weight assessment ratio Analysis (SWARA) and complex proportional assessment of alternatives with grey relations (COPRAS-G), were applied for market segment evaluation and selection. The most desirable features influencing the choice of a market segment evaluation and selection are identified based on literature study. Grey relation analysis allows incorporating the vague and imprecise information in to the decision model. A real-world data on a laptop market is put forward to illustrate the performance of the proposed methodology. The proposed model could help companies to segment, evaluate and select the best market.Keywords: Market segmentation, Market segment evaluation and selection, Data mining, Clustering, MADM, SWARA, COPRAS-GIntroductionNowadays, the business environment is more and more competitive and customers are highly demanding (Aghdaie et al., 2011). Market consists of variety types of customers with different characteristics, needs and wants so companies cannot satisfy the whole market. For dealing with these problems market segmentation approaches have been used by many companies. The idea of market segmentation was introduced by Smith (1956) into the academic marketing literature, however, after about fifty years this idea goes on to be a critical focal point of ongoing research and marketing practices (Chaturvedi et al.,1997). According to Kotler (1999) market segmentation is the partitioning of a market into distinct subsets of customers and any subset could be possibly selected as a target market to be reached with a distinct marketing mix. Kotler (2001) suggested general bases in marketing literature which are proposed to divide a heterogeneous market into homogenous subsets of segments such as geographic, demographic, psychographic and behavioral.Decision making about market segmentation, evaluation and selection is based on data or information which is collected from customers, competitors and other parts of market. Recent developments of information technology and internet era accelerate collection of vast data for business decisions. Data has become a critical resource in many organizations, and therefore, efficient access to data, sharing the data, extracting information from the data, and making use of the information has become an urgent need (chiu et al., 2009). According to Turban et al. (2007) data mining is a process which uses statistical, mathematical, artificial intelligence and machine-learning techniques to extract and identify useful information and subsequently gain knowledge from large databases. Data mining techniques has been used frequently for market segmentation studies (Liu and Chen, 2009) (Fathian and Amiri, 2008) (Cheng et al., 2005) (Rafalski, 2002) (Hanafizadeh and Mirzazadeh, 2011) (Kazemzadeh et al., 2009).After segmenting a market, another important problem for every organization is evaluation and selection of a segment or segments as a target market. Although much of the marketing literature has proposed various market segmentation techniques, but a review of academic research reveals that existing studies have relatively neglected segment evaluation and selection (Sarabia, 1996; Ou et al., 2009; Aghdaie et al., 2013). Selecting an appropriate market segment based on evaluation of segments is one of the most complicated and time consuming problems for many companies, due to many feasible alternatives, conflicting objectives and variety of factors (Aghdaie et al.,2011). Market segment evaluation and selection decisions are sophisticated by the fact that the decision-making process must consider various criteria. Therefore market segment evaluation and selection can be viewed as a multiple attribute decision making (MADM) problem. The MADM methods deal with the process of making decisions for finding the optimum alternative in the presence of multiple, usually conflicting, decision criteria. Hence, this study has the main objective of proposing a mechanism for market segmentation, segment evaluation and selection. More precisely, in this paper we proposes to use the technique of data mining combining with MADM tools for market segmentation of customers and then evaluation and selection of the best segment or segments as a target market.Through this study, we expect to build an effective and accurate market segmentation system which in this system market segmentation, evaluation and selection of the best market managesin sequence.The remainder of this paper is organized as follows. In the next section, a literature review of market segmentation, data mining and MADM topics are illustrated in details. Section 3 outlines the proposed methodology combining, Clustering, SWARA, and COPRAS-G. In Section 4, a real-world data is given to prove the applicability of the proposed method on a laptop market in Iran. Also in Section 4, the results are discussed. In Section 5, finally, the article will be concluded and future researches are discussed.2. Literature review2.1 Market segmentation and market segment evaluation and selectionMarket segmentation idea was first introduced by Smith (1956), an American marketing scientist, and this concept further followed as an approach by many companies and practitioners. Now a large number of companies or organizations are available to describe about the number of achievements which were gained by this approach.A crucial and common issue with market segmentation for both academicians and practitioners is how best to subdivide a market. A review of the marketing literature shows that there is no one correct way to segment a market (Beane and Ennis, 1987; Kotler et al., 2010). It has been argued that different approaches can be used to satisfy the researchers’ requirements (Tkaczynski and Rundle-Thiele, 2010). According to Kotler (1980) the most popular segmentation approach is a combination of the segmentation bases.Another important marketing point after market segmentation is that how a company can evaluate and select the best market segment or segments to focus its own effort on it. According to Weinstein (2004) companies must carefully assess and weigh key discriminating criteria to find the “best” market segments. A review of related studies indicates that there are a few studies which focus on market segment evaluation and selection. Even general studies of market segmentation have paid little or no attention to the evaluation and selection stages (Beane and Ennis, 1987; Weinstein, 1987; Wind, 1978). Besides, the major part of the related literature concentrates on the important features for doing this evaluation and very little research has been done on the evaluation of segment attractiveness and market segment selection. Furthermore, in these studies authors usually limit themselves to analyzing how to evaluate segment stability (Bettman, 1971; Calentone and Sawyer, 1978; Lehmann et al., 1982; MacLachlan and Johansson, 1981), congruence (Green, 1977), internal homogeneity and profitability (Eckrich, 1984; Van Auken and Lonial, 1984; Beik and Buzby, 1973), to mention only the most relevant.Table 1 depicts the market segment evaluation criteria which are used by different authors including profitability, market growth, market size, identifyability, substantiality, accessibility, stability, responsiveness, actionability, variability, accessibility, profitability, reachability,measurable, sustainable, differentiable, homogeneity, defensibility, competitiveness, durability, compatibility and risk.Table 1: market segment evaluation criteriaAuthors CriteriaFrank et al, 1972; Loudon and Della Bitta, 1984; Baker, 1988; Kotler, 1988 identifyability, substantiality, accessibility, stability, responsiveness, actionabilitySimkin and Dibb (1998) profitability, market growth, market sizeMcQueen and Miller (1985) profitability, variability, accessibilityLoker and Perdue (1992) profitability, accessibility, reachabilityKotler and Armstrong (2003) measurable, accessible, sustainable, differentiable, andactionableMorrison (2002) homogeneity, defensibility, competitiveness, durability, andcompatibility.Jang et al. (2002) Profitability, risk, risk-adjusted profitability and relativessegment sizeLee et al. (2006) Evaluating profitability (using the Economic Value PortfolioMatrix and market size)In addition, McDonald and Dunbar (2004) prepared a comprehensive criteria list for market segment evaluation. They also provide a list of twenty-seven possible, generalized segment attractiveness factors in five major areas: segment factors, competition, financial and economic factors, technology, and sociopolitical factors. Besides Lu (2003) has evaluated international distribution centers based on shippers’ service requirements. In this study factors which are related to distribution centers such as cargo safety, cargo tracing service, inland transportation, and customs clearance was selected for market segment evaluation. Furthermore, Porter (1979) introduced an idea of five forces analysis and this model is explained in detail in his book “Competitive Strategy” (Porter, 1980). Ou et al.(2009) used this model to evaluate each potential segment.2.2 Data miningData mining (DM) has recently seen an explosion of interest in many fields of applications, owing to the increasing amount of data available, and the growing understanding that deeper analyzes are far more valuable than simple summary statistics (Corne et al., 2012). According to Fayyad (1996) Data mining is a non-trivial process of identifying valid, novel, potentially useful and ultimately understandable patterns in data. Data mining enables businesses to extract hidden information from large amounts of data so that they can better understand their consumers so data mining interest and application are increasing (Chopoorian et al.,2001). The explosive growth in databases has created a need to develop technologies that use information and knowledge intelligently (Liao et al., 2012).According to (Han and Kamber, 2001) some of the most important functions of data mining include concept description (characterization and discrimination), association, classification, clustering, and prediction.2.2.1 Clustering techniquesAlthough there are a lot of market segmentation techniques in the marketing literature, clustering techniques are commonly used in practice (Wedel and Kamakura, 2001). Clustering is a useful technique for the discovery of some knowledg e from a dataset and it’s an exploratory method for helping to solve classification problems (Chiu et al.,2009). This technique is widely used in customer segmentation (Chiu and Tavella, 2008; Dillon et al.,1993). Also, it’s a convenient method commonly used for the identification and definition of market segments (Hong, 2012). Besides, according to Peacock (1998) clustering is a fundamental data mining task for segmentation.Hierarchical and partitional are two major groups of cluster analysis (Jain and Dubes, 1988). In another view clustering methods can be classify into hierarchical and non-hierarchical groups. Previous researchers have noted that there are over 50 clustering methods in marketing literature and these methods has been applied to market segmentation problems (Milligan and Cooper, 1985) .In clustering problems, K-Means algorithm which is developed by MacQueen (1967), is one of the most popular, simple and frequently used algorithms (Hung and Tsai 2008). K-means algorithm is a famous clustering algorithm which belongs to non-hierarchical group. Though K-means was first proposed over 50 years ago, it is still one of the most widely used algorithms for clustering (Anil, 2010). The main reasons for its popularity are ease of implementation, simplicity, quick, efficiency, and empirical success (Forgy, 1965; Mirkin, 2005). Besides, it can accommodate the large sample sizes associated with market segmentation studies (Anil et al., 1997). One of the disadvantages of the k-means algorithm is that the number of clusters must be supplied as a parameter (Gan et al., 2007). For identifying the number of clusters which K-means algorithm is required, Ward’s method was used. This integration forms a new clustering method which was called two-stage clustering method. Punj and Steward (1983) suggested that this integration is a feasible solution for clustering. The reason is that hierarchical methods, like Ward's minimum variance method, can determine the candidate number of clusters and starting point that non-hierarchical methods, like the K-means method, need, while non-hierarchical methods can provide better performance with the specified information (Kuo et al.,2002). Ward’s minimum variance was used because it has been work ed well in earlier studies (Hair et al., 1995; Malhotra, 1993). Also, this method can do clustering appropriately with ordinal data (Everitt, 1993). In addition, Ward’s (1963) method achieves better results with respect to other hierarchical clustering methods except in the presence of outliers (Sharma, 1996; Punj and Steward, 1983). Ward’s (1963) method is one of the best and most popular hierarchical clustering techniques (Mingoti and Lima, 2006). Furthermore, this method tends to create segments such that the variation within these segments does not increase too radically (Hardle and Simar, 2003).2.3 MADMOperations research/Management science has many sub-disciplines which MADM is one of them. Also, MADM is one of the two main categories of multi criteria decision making (MCDM). The other category of MCDM is multi objective decision making (MODM). Nowadays, the number of publications in the field of MADM has been grown rapidly and many researchers have used this approach for solving their problems. This approach has been used in many fields for decision making including healthcare, industries, tourism, energy, supply chain management, marketing, finance, human resource management, strategy, militarily, agriculture, fishing and so on. There are some reasons for this explosive growth.(1) Nowadays, Decision maker (DM) should deal with various issues such as fast changing business environment, less predictable sophisticated and dynamic situations, tough competitions, risk and uncertainties, finite and infinite information. Also, DM should assess a set of alternatives or scenarios with respect to numerous usually conflict and multiple criteria for selecting the best decision.(2) MADM approach makes a decision more visible to others, provides a focus for discussion, supplies a system of problem structuring and working through the information, and breaking the decision down so that people better understand the decision both from th eir own and from others’ perspectives (Belton and Stewart, 2002)(3) Developments of personal computers, related computer software and internet.(4) Ability to incorporate with fuzzy logic or grey relation analysis for dealing with imprecise and vague information.(5) Heuristics become an important approach for solving problems in recent years and they can easily use by MADM methods (Wallenius et al., 2008).(6) Ability to participate number of experts for group decision making (Dooley et al., 2010).(7) Flexibility and adaptability to incorporate with other operations research methods.Some of the famous MADM methods in the literature include analytic hierarchy process (AHP) (Saaty, 1980), fuzzy analytic hierarchy process (Torfi et al., 2010), analytic network process (ANP) (Saaty and Vargas, 2001), technique for order preference by similarity to ideal solution (TOPSIS) (Hwang and Yoon, 1981; Isiklar and Büyüközkan, 2007), Elimination and Choice Translating Reality (ELECTRE) (Roy 1968; Roy 1991),MUSA (Grigoroudis and Siskos, 2002), AKUTA (Bous et al., 2010), VIsekriterijumska optimizacija i KOmpromisno Resenje (VIKOR) (Opricovic, 1998), fuzzy VIKOR (Chen and Wang, 2009), Preference Ranking Organization Method for Enrichment Evaluations (PROMETHEE) (Brans et al.,1984; Brans and Vincke, 1985), Simple Additive Weighting (SAW) (Churchman and Ackoff, 1954, COmplex PRoportional ASsessment (COPRAS) (Zavadskas and Kaklauskas, 1996; Zavadskas et al., 2007) COmplex PRoportional ASsessment with Grey relations (COPRAS-G) (Zavadskas et al., 2009), Step-wise Weight Assessment Ratio Analysis (SWARA) (Keršulienėet al., 2010), Factor Relationship (FARE) (Ginevicius, 2011).3. MethodologyThis section describes a two phase approach which is used for market segmentation and market segment evaluation and selection. In this conceptual model two methodologies have been combined (see Fig 1). Clustering as a famous data mining technique was used for segmentationwhich market segmentation is based on its results in phase I. Phase I is market segmentationphase and for conducting this phase five steps should carry out.Fig1: Market segmentation and market segment evaluation and selection procedureMADM techniques were used for evaluation and selection of generated clusters as potentialsegments. Phase II, is market segment evaluation and selection phase. For conducting this phase,eight steps should carry out. The most important criteria for market segment evaluation andselection was identified. Next, the qualitative and quantitative criteria were identified. Thecriteria listed in Table 2 are selected based on the literature survey. Finally, the project team constructed the selection criteria and problem structure (see Figure 2). Criteria weights wascalculated by applying SWARA method and based on experts ‘evaluations. In this stage, all alternatives were evaluated by a group of experts and COPRAS-G method wasapplied to achieve the final ranking results.Table 2: Factors taken from the review of the related literature which are relevant to market segment evaluation andselectionNo. Criteria Sub-criteria Related literature sourceX1Segment relatedX1-1-Growth rate per year Lee et al. (2006); Proctor (2005);Simkin and Dibb (1998)X1-2-SizeMcDonald and Dunbar (2004); Simkin and Dibb (1998); Aghdaieet al. 2011X1-3-Suppliers ability Morrison (2002); McDonald and Dunbar (2004); Aghdaie et al.2011X1-4-Homogeneity Aghdaie et al. (2011); Kotler and Armstrong (2003); Loker andPerdue (1992)X1-5-Accessible Morrison (2002)X1-6-Sustainable Kotler and Armstrong (2003)X1-7-Competition McDonald and Dunbar (2004);Ou et al (2009)X2Financial and economicX2-1-ProfitabilityMcQueen and Miller (1985);Loker and Perdue (1992); Jang etal. (2002); Simkin and Dibb(1998)X2-2-Risk Jang et al. (2002) X3TechnologicalX3-1-Knowledge, experience,information, and manufacturingprocess technology requiredMcDonald and Dunbar (2004);Kotler and Armstrong (2003)X3-2-Complexity McDonald and Dunbar (2004) X4Socio-politicalX4-1-Social attitudes and trends McDonald and Dunbar (2004)X4-2-Laws and governmentagency regulationsMcDonald and Dunbar (2004)Figure 2: Market segment evaluation and selection structure, selection criteria and alternatives3.1 Data sourceSurvey data were collected through questionnaires, between March 1 and March 3, 2011. The total sample size consisted of 760 respondents, 230 women and 530 men. All respondents are Iranian, aged 13 years old or older, who tends to buy laptop soon. The questionnaires collected a wide range of information including socio-demographic characteristics and behavioral (see appendix 1). The age of respondents range from 13 to 55 years, with a mean of 26.54 (SD = 6.95). For evaluations of criteria in questionnaire a rating system was designed. The ratings were obtained using ten-point scale (1= Least preferred, 10 = Most preferred).The questionnaires distributed in Paytakht Shopping Center. Paytakht Shopping Center is the biggest and primary shopping center for selling and buying laptops. Besides, this place is a hub of laptop computers which is located in capital of Iran, Tehran. The main reason for selecting this center is that about 70% percent of total laptops which is sold in Iran, has been sold through this center. This center is always crowded with customers and many buyers form other cities that come there for buying new laptops.3.2 ClusteringIn this study, two different types of cluster analysis techniques is used to cluster customers based on their main benefits into meaningful homogeneous sub-groups which may exist within in the market. Firstly, Ward’s method is used to determine the optimum number of clusters. Secondly, the K-means clustering technique was applied for final clustering with the initial number of clustering. Although Ward’s method has performed successfully in some of the earlier studies, but non-hierarchical method are superior to hierarchical methods (Punj and Stewart, 1983). Theyare more robust to outliers and the presence of irrelevant attributes (Wedel and Kamakura, 2001).A large numbers of non-hierarchical methods are available but K-means is the best known and most widely used of those procedures (Wedel and Kamakura, 2001). Punj and steward (1983) suggested that integration of Ward’s method and K-means method can provide better results for clustering. The most important reason for such integration is that Ward’s method can provide the optimal number of clusters and starting point of each cluster which the K-means method requires to determine the final solution due to its efficiency. K-means method with a derived commonly performs better than other methods across all conditions and provides the best recovery of cluster structure (Punj and Stewart, 1983). Also among clustering methods, the K-means method is the most frequently used, since it can accommodate the large sample sizes associated with market segmentation studies (Anil et al., 1997).3.3 A step-wise weight assessment ratio analysis (SWARA) method (Keršuliene et al. 2010) Academics has been suggested a variety of approaches for assessing weights such as (Zavadskas et al.2010a, b), e.g. the eigenvector method, SWARA (Keršuliene et al., 2010), expert method (Zavadskas, Vilutienė, 2006), analytic hierarchy process (AHP) (Saaty, 1977; 1980), FARE (Ginevicius, 2011). The step-wise weight assessment ratio analysis (SWARA) method which is developed by Keršuliene et al., 2010 is one of the brand-new MADM methods. In this method, the most significant criterion is given rank 1, and the least significant criterion is given rank last. The overall ranks to the group of experts are determined according to the mediocre value of ranks (Kersuliene and Turskis, 2011). Also, this methodology has applied for the selection of rational dispute resolution method (Kersuliene and Turskis, 2011). The procedure for the criteria weights determination is presented in Fig. 3.The main feature of SWARA method is the possibility to estimate experts or interest groups opinion about significance ratio of the criteria in the process of their weights determination (Kersuliene et al, 2010). This method is useful for coordinating and gathering data from experts. Besides, SWARA applications are uncomplicated and experts in various fields can contact with general idea of this method easily. The all developments of decision making models based on SWARA method up to now are Keršuliene et al. (2010) in selection of rational dispute resolution method, Kersuliene and Turskis (2011) for architect selection, Hashemkhani Zolfani et al. (2012c) in design of products.Fig 3: Determining of the criteria weights based on (Kersuliene and Turskis, 2011)3.4 The COPRAS-G methodFor choosing an alternative among other alternatives, alternatives should be evaluated by DM with respect to criteria. In order to evaluate the overall efficiency of an alternative, it is necessary to construct a problem structure and selection criteria. Then DM should assess information,relating to these criteria and finally DM should develop methods for evaluating the criteria to meet the participants ‘needs. Complex proportional assessment (COPRAS) method that was developed by Zavadskas and Kaklauskas (1996) is one of the MADM methods. This method was applied for evaluation of alternatives in many MADM problems. In real decision making situations, the most alternatives always deal with vague information, and evaluations of criteria expressed with linguistic terms. Also, it’s hard for DM to describe evaluations with precise numbers. Therefore MADM approaches should be functioned not only with exact criteria values, but with fuzzy values or with values in some intervals. For dealing with mentioned reasons, Zavadskas et al. (2008) presented the main ideas of complex proportional assessment method with grey interval numbers (COPRAS-G) method. The idea of COPRAS-G method with criterion values expressed in intervals is based on the real conditions of decision making and applications of the Grey systems theory (Deng 1982; 1988).The recent developments based on COPRAS-G method are listed below:- Hashemkhani Zolfani et al. (2011) in forest roads locating using COPRAS-G method. - Hashemkhani Zolfani et al. (2012a) in supplier selection using COPRAS-G method. - Hashemkhani Zolfani et al. (2012b) in quality control manager selection applying COPRAS-G method.- Aghdaie et al. (2012a) prioritizing projects of municipality applying COPRAS-G method.- Rezaeiniya et al. (2012) greenhouse locating using COPRAS-G method.- Bitarafan et al. (2012) Evaluating the construction methods of cold-formed steel structures in reconstructing the areas damaged in natural crises based on COPRAS-G. The procedure of applying the COPRAS-G method consists in the following steps (Zavadskas et al., 2009):1. Selecting the set of the most important criteria, describing the alternatives.2. Constructing the decision-making matrix X ⊗:[][][][][][][][][][][][][][][]1111111121211212212122222211122;;;;;;;1,,1,;;;m m m m m m n nm n n n n nm nm x x x x x x x x x x x x x x x x x x x x x x x x X j n i m ⎡⎤⎡⎤⊗⊗⎢⎥⎢⎥⊗⊗⎢⎥⎢⎥⊗====⎢⎥⎢⎥⎢⎥⎢⎥⊗⊗⎢⎥⎢⎥⎣⎦⎣⎦(5)Here ji x ⊗ is determined ji x (the smallest value, the lower limit) and ji x (the biggest value, the upper limit).3. Determining significances of the criteria i q .4. Normalizing the decision-making matrix X ⊗ are calculated by formula 6:,221~1111⎪⎪⎭⎫ ⎝⎛+=⎪⎪⎭⎫ ⎝⎛+=∑∑∑∑====n j ji n j ji ji n j n j ji ji ji x x x x x x x ;)(221~111∑∑∑===+=⎪⎪⎭⎫ ⎝⎛+=n j ji ji jin j n j ji ji ji x x x x x x x(6)。