K-means and SOMs in Capturing and Evaluating Markets - Research Proposal Example

K-means and SOMs in Capturing and Evaluating Markets (Name) (University) Abstract Market segmentation is imperative for an organization’s marketing given that the it offers an important platform for effective marketing strategies. In the contemporary business environment, due to their increased performance in artificial neural networks, has been integrated in management approaches. Otherwise known as SOMs, Self organizing maps, an approach of unsupervised neural networks, are used for grouping. This retrospect study seeks to integrate a two stage strategy, K-means and SOMs, to effectively group consumers and provide a comparison in performance between k-means and tandem approach, and test for the presence of segments and clusters. It is evident from the tests that the media promotion aspect can be effective basis for segmentation. With respect to the obtained results therefore, a marketing communication strategy is created in coping with consumer expectation. List of Figures Table 1: Fourteen Benefit Variables Used Table 2: Results From Regression tests Table 3: Frequencies Figure 1: Cluster visualization for training samples in potential markets Figure 2: Cluster visualization for testing samples in potential markets. Contents 1.0 Introduction 5 2.0 Literature Review 6 2.1 SOMs in market segmentation 6 2.2 Test of Clusters and Segments 8 3.0 Methodology 9 3.1 Research sample 9 4.0 Conclusion 19 5.0 References 19 K-means and SOMs in Capturing and Evaluating Markets 1.0 Introduction From the time that the new methods of Wendell Smith (1956), the notion of segmenting markets remains as one of the most permeating procedures in two areas; academic marketing literature and practice (Kuo, Ho, and Hu, 2002).In recent competitive world of commerce, placing and sufficiently targeting one of a kind market partitions allows an organization to comprehend the preferences of its clients. For a few decades, analysis of statistical clusters has been fruitfully utilized in market partitioning, owing to increased levels of computer power, and a decline in the cost of computer, a greater amount of involvement and exertion have been positioned in the direction of neural networks (NNs) in business practices initially set aside for the statistical analysis of multivariate (Madhulatha, 2012). NNs is the main application on market partitioning in the marketing sector. Self-organizing maps, with the evolution of new data mining methods, have been used in ascertation of clusters as a possible approach in statistical clustering methods (Subasic and Klobucar, 2012; Rönnqvist and Sarlin, 2013; Kohonen, 2014; Aydin et al., 2016; Kuo et al, 2002).A SOM furnishes a correspondence from an elevated into the lower dimensional input data production maps. Guarding of the input data properties from an elevated form of dimensional input volume in such a manner that comparative measures connecting the input data are approximately preserved on the output representation is one of the distinct structures of the SOM (Aydin et al., 2016, Thanakorn 2003). In spite of a number of grouping techniques having been shown to work out the market partitioning issue, the significance of trialing the authenticity of groups is often neglected by market researchers (Seward and Doane, 2011; Engelman & Hartigan 1969,Pilling,Crosby & Ellen 1991,Punj & Steward,1983).If established,groups are insufficiently authenticated, such grouping conclusions can be regarded as random. Groups are reffered to as clusters showing some models of resemblance and that are different most importantly with regards to their reaction to the marketing variables applicable (Sommers & Barnes 2003).Groups established an approach that to diverge someway, for example attitudes or features unrelated to behavior are do not make up real portions (Massy & Frank,1965). Kuo et al, ( 2002) suggested a partially changed double staged approach (SOMs and K-means) and focused mainly on stimulated ideas, they discovered that it had an advantage over the old fashioned statistical clustering method. This research uses applicable facts sourced from local heating organized market that attaches Kuo ET ales two- stage method in market partitioning, differentiates execution of straight K-means grouping with the tandem approach. 2.0 Literature Review 2.1 SOMs in market segmentation The clustering of homogenous items is the prevalent essence of partitioning (Kuo et al 2002, Smith 1956).Initially, market partitioning has seen statistical grouping methods as the frequent equipment (Madhulatha, 2012). Seward and Doane (2011) recommended that the incorporation of a hierarchical approach such as Wards minimum wage variance, in line with a non-hierarchical one similar to the K-means that had the capability to furnish an improved solution than making use of either of the two; ranked or non-hierarchical approach solely. This approach is the commonly known as the two-stage approach. The phrase virtual computer systems came from unnatural capacity study which made an effort to comprehend and design the behavior of the brain (Berry & Linoff 2000,Berson,Smith and Thearling,2000).Self arranging representations are feed-forward, lack supervision of virtual computer systems (Kohonen 2001). Networks based on the feed-forward system are utilized in circumstances such that they are able to generate all of the knowledge to sustain on a difficulty at once (Seward and Doane, 2011).The SOM comprises of two layers; processing units and an input layer that is in full connection to an output layer. No hidden layer exists. The unobserved virtual computer systems including SOMs and the adaptive resonance theory, was opposed since it could be used in establishing groups. Their structure for partitioning was imaginally created. Fish et al(1995) also implied that for market partitioning, unobserved virtual computer systems could be used. With the accomodation market, Kuako , Hoomineijer & Hakfoort (2002) evaluated the use of virtual computer systems to accomodation market partitioning in Finland. Their research outlines the inevitability of establishing several aspects of the housing partitions by unmasking models in the data set at the same time grouping skill of SOM-LVQ (Learning vector quantization).Kuo et al (2002) based on Punj & Steward (1983) research suggested a partially changed two-stage approach that involved using self-arranged representations to establish how many groups were involved and proceed to implement the K-means technique and arrive to the last answer. Kuo et al (2002) made use of replicated data and discovered that the two-stage approach they suggested performed better than the original two-stage method. The principal reason was the association to hierarchical methods of the first stage of the ordinary two-stage grouping technique. Non-recovery component was the first challenge encountered. No repositioning of a notion once designated to a group was to take place all (Kuo et al 2002).Nonetheless, SOMs are a kind of educational algorithm capable of performing a repeated update or reassign the observation to the nearest group. In Kuo et al., research, they implemented the tandem method was faulted in that it could disfigure the initial group form (Madhulatha, 2012). Evaluation of the non-tandem approach is paramount and examination of its accomplishment for example the direct K-means grouping. 2.2 Test of Clusters and Segments A suitable evaluation for groups would be one that puts into consideration the group analysis goal i.e. to reduce the within the variance of the group (Seward & Doane, 2011).Among the examining techniques, a technique recommended by Arnold seems to be improved than other techniques (Seward and Doane, 2011).In Arnolds exam, the importance of the solution of the group is evaluated by illustrating the value of the C test quantities previously computed to the expected values supposing the facts were derived from either of two distributions; unimodal or a uniform i.e. zero basis for groups (Seward and Doane, 2011). There is a general agreement in the marketing research on the quality of a partitioning answer as a good one (Bacon 2002,Wedel & Kamakura,2000). Accepted crucial standards are;associability,receptibility,possesibility,stability,answerable and action ability (Bacon 2002,Wedel & Kamakura 2000).These standards are with respect to service for managerial judgement and are difficult to examine (Bacon 2002).These standards are devoid of the technology used to acquire them (Bacon 2002). They therefore furnish an important pack of objectives for portion solutions created using virtual computer systems methods (Bacon 2002).Several researches tried to evaluate portioning answers using this method (Wedel & Kamakura 2000). One criterion-responsiveness are commonly used in this kind of paper. The main reason for selecting responsiveness is that the information used was sampled. Other methods e.g. stability, need data sampled over time. Responsiveness is in close relation with Massy & Franks (1965) approach; distinct portions need assorted ranking or marketing differences elastic. Statement of the Proposed Research This retrospect study seeks to integrate a two stage approach, K-means and SOMs, to effectively group consumers and provide a comparison in performance between k-means and tandem approach, and test for the presence of segments and clusters. 3.0 Methodology 3.1 Research sample This study used to collect data for analysis grouping. Data that were gathered are sample facts from the consumer’s market panel mail in 1983. Due to the freezing weather cold season heating system at home are vital in some states, for instances state of New York. By nineteen-eighties, there were around five types of heating system; such home heating system includes oil, gas, electric, solar as well as a heat pump. Families chose only one type of heating system in order to use it for warming the house. Through some research done by the Haines and Ratchford, they discovered fourteen perpetual attributed by the criteria of rating five heating systems type at home. The researchers also used analysis of factoring among the identified characteristics, as they noticed that the reliability, cleanness as well as cost are the resulting factors that explain approximately sixty percent of the different attributes but unfortunately they did not operate the segmentation among the users. Consumers planning is what is focused when purchasing a house in the next coming five years. The research explains the respondent with no plan of replacing the home heating system shortly probably five years. The pan of purchasing house in the next five years has a potential market. One must consider the type of heating system in a house before planning to purchase it. Both male and female were included in the subject, and more than eighteen point four three respondents were obtained from the market. Research hypotheses Clusters testing From the groups testing procedures that were described by the Arnold depict two sets. The initial sets are the null hypothesis test which indicates the population stages that tends to concentrate at single point for instance single normal distribution against the hypothesis that are alternative which the objects can be distributed uniformly, or it can be categorized into groups. Then, when the hypothesis is null, then it cannot be accepted at various confidence level, which results to test two be made. Testing two indicates that the resulting hypothesis is null, and the population entities are distributed uniformly against the alternative hypothesis which forms two and more groups. The hypothesis that are null include, H1, population bodies tends to concentrate at single point. H2, the stages of the population are distributed uniform Rejecting H1 as well as H2 shows clusters in the population. Segments testing Null hypothesis indicates that: H3: show market elasticity of variables among the clusters being the same. Rejecting H3: indicates that the clusters are segmented. Questionnaire Feign, young and Ott explained that segments that bases on the profits desires is always significant categories to use at standpoint marketing when it facilitates product directly through positioning as well as communicating during advertisement. The backbone point of the benefits segments benefit soughed by the consumers will tend to be causal relation to future buying behavior. The study uses variables of benefit to group customers. The questionnaire elaborates news on five ctegories of home heating system with every system rated at fourteen attributes. The benefits are Table below indicates fourteen variables benefits Reliable against breaking down Availability of future fuel supply The system requires floor space For converting fuel source efficiently to heat The system requires cleanliness before operating. Helps in converting one system to another There is no fumes and odors Results to avoiding pollution Professional services are available There is no noise from the operating system the system safety is considered The original buying price is indicated Protecting warrant is issued Evaluation of the operating cost every year Clusters generation using SOMs Different steps can be used to summarize Kohonen algorithm as well as self-organizing map parameters used in the study First, the weight of a neuron is initialized to random values. Secondly, data is represented. When data is clustered over neural networks, standardized practices normalizes the data input to range of zero to one. Sigmoid function is used in order to transfer data to range of zero. Thirdly, learn rate setting. The used rate controlling the stage size in the adjustment connection weight. Fourthly, neighborhood setting. The output layer be ten by ten. Fifth, presenting the training cycle number. The training cycle is a data set training to presents model among the neural weight updates Sixth, the data input at the net point are selected randomly Table below indicates parameters ofself-organizing maps The parameters of self-organizing maps include (0) Input Unit numbers are ( Read More