International Journal of Computer Applications (0975 – 8887) Volume 106 – No.10, November 2014 5 Marketing Data Mining Classifiers: Criteria Selection Issues in Customer Segmentation Masoud Abessi Department of Industrial Engineering, Yazd University, Iran Elahe Hajigol Yazdi Department of Industrial Engineering, Yazd University, Iran ABSTRACT Data mining is automated or semi-automated Knowledge Discovery from large amounts of stored data in order to discovering meaningful patterns and rules. Marketing related data mining applied to market segmentation, customer services, credit and behavior scoring, and benchmarking. There are different classifiers including decision tree, ID3, CART, Quest, Neural networks, Association, Bayesian, and etc. In this study, ten classifiers are examined to identify important issues in mining marketing data. Classification accuracy, learning speed, Classification speed, Missing value, and robustness are some indices included to compare and contrast the classifiers. Shopping malls’ consumer behavior data were used in our investigation. Results indicate that classifiers perform differently under different consumer data types. Keywords Data mining, Supervised learning algorithm, classification techniques 1. INTRODUCTION Knowledge discovery from marketing data is an increasingly important phenomenon. Classification techniques can learn from consumer behavior, market segmentation, and customer purchasing habit to promote Customer loyalty and profitability, and products/services Customization (Gloy, Akridge, & Preckel, 1997). Marketing classifiers in a supervised learning algorithm format helps to discover distinctive groups based on certain preferences. In this study, we examine different classifiers to identify various market segmentation based on observed customer purchasing and product usage behavior, socioeconomic factors, demographic, and life stage characteristics. It creates specific opportunities to plan a strategic roadmap which` leads to the greater customer loyalty, shorter customer purchase cycles, higher spend, and lowering service and support costs (Garofalakis et al, 2000). Various segmentation techniques exist to classify costumers according to their unique characteristics, each with their own unique advantages and challenges. Segmentation is a multi- disciplinary method which combines statistics and machine learning techniques. Statistic approach is implemented for collecting, classifying, summarizing, organizing, analyzing, and interpreting data. Statistical approaches like regression classifier can’t extract tacit knowledge which is hidden in datasets. Legitimately, they cannot generate rules. Therefore, this approach doesn’t have learning ability. Various studies were implemented statistical segmentation methods. Gilboa (2009) presented a two-step cluster analysis based on customer’s socio-demographic characteristics to segment Israeli mall customers as disloyal customers, family bonders, minimalists, and mall enthusiasts. Wu (2006) implemented factor analysis and Discriminant classifier based on the lifestyle and personality characteristic Factors to predict the intentions and behavior of different Internet bookstores customer groups. Regression analysis applied to figure out the relationship between service quality perceptions and satisfaction and intention by lee et al (2011). On the other hand, data mining techniques exemplary machine learning methods act based on finding structure in the dataset. Kim et al. (2006) utilized decision tree with the aim of customer’s identification. Whereas, target customer analysis take placed by using decision tree (Chen et.al, 2003; Wu et.al, 2005). Liang (2010) analyzed customer value by data mining technologies Integration as SOM and decision tree for the automotive maintenance industry. According to the literature review, vast approaches were implemented for costumers’ segmentation without any harmony among them. Most studies sufficed to statistical analysis of customer’s data which learning process didn’t take place by it. They explore data by means of describing it based on its important aspects. Hidden knowledge in data set didn’t extracted; therefore, predictive power of statistical segmentation methods is low. Studies which used DM techniques for customer grouping often have a special attention to results achieved from segmentation and its characteristics. They didn’t attend to model deployment according to learning process, predictive power, and knowledge tacit in extracted rules. The results of the KDnuggets 2009 Data Mining Deployment Poll indicate that more than 55% of people who used machine learning techniques didn’t deploy the model and just utilized the results to gain business, scientific knowledge, or publish papers. Various Learning methods work under their unique conditions. Another problem with segmentation model selection is that the customers data are normally stored in various forms including binary values (asymmetric or symmetric), ordinal or ratio scales depending on marketing subjects. For example many customers’ data are collected in Likert scale or other ordinal forms from questionnaires filled by consumers (Michael & Gordon, 1997). Whereas, different validation techniques are implemented various evaluation criteria. They are including Predictive (Classification) accuracy, speed, robustness, Scalability, Interpretability, Simplicity and Domain-dependent quality indicators (Yeh &Lien, 2009; Ture & et.al, 2009). Therefore, the admissible classification method varies according to the data characteristics and business requirements (Carrier & Povel, 2003). This paper aims to represent a framework to facilitate classifier selection for customer segmentation according to data characteristics and business requirements, legitimately. First, classification models and their characteristics presented. We refer to their strengths and limitations (Kims 2008). Critical factors in classification method selection which is applicable to marketing data proposed for ambiguous reduction in technique selection. We apply various classification models to a shopping malls data set to classify shopping malls customers according to their demographic and