Revisiting the problem of market segmentation: A new approach using self -organizing maps, a data mining technique, in database marketing

In today's competitive marketplace, locating and effectively targeting unique market segments is both a reality and a necessity. For the implementation of segmentation, the choice of clustering techniques and segmentation base, the test of clusters, and the evaluation of segments are important factors for the success of segmentation.

Due to the increasing popularity of data mining, the study explores data mining from a business perspective, and clearly discusses the data mining definition, process, functions, techniques, business applications, and approach. Through the assessment of data mining, the study examines the current limitations and challenges to data mining.

Self-organizing maps (SOMs), a data mining technique, are often used for clustering or dimensional reduction. Data mining is a way to analyze an enormous amount of data to find trends or relationships that were not previously known to exist. The study focuses on the segmentation issue and applies direct K-means clustering in the modified two-stage approach (SOMs and K-means) to group customers. Its performance is proved to be better than tandem approach. The process of systematically collecting data and using data mining or other statistical techniques to transform data into information and to formulate marketing strategy is called database marketing (DBM).

This research reviews the marketing literature and redefines DBM. It also clarifies the relationships among direct marketing, relationship marketing, and database marketing. The study finds database marketing is evolved from direct marketing (but DBM is not direct marketing), and argues when DBM is used for a strategic purpose, it can lead to developing relationship marketing programs.

Based on the importance of segmentation, the author employs benefit variables as segmentation bases and uses previously collected data from Ratchford and Haines (1986). After clustering the customers, a test of clusters (Arnold, 1979) shows the clusters exist in the target market and potential market. But a following test of segments (Massy and Frank, 1965) indicates the clusters are not segments in both the target and potential markets. However, this research further examines significant regression coefficients in cluster one and cluster two in the potential market, and the result reveals that media promotion variable is significant and it would be a basis for segmentation. Finally, marketing segmentation strategies are presented to cope with customers in the two segments.