Market Segmentation - Algorithms and Approaches

Algorithms and Approaches

Any discrete variable is a segmentation. For instance, customers might be segmented by gender ('Male' or 'Female') or attitudes ('progressive' or 'conservative'). Numeric variables may be discretized to become segmentations, such as age ("<30" or ">=30") or income ("The 99% (AGI= US $300,000)").

Segmentations can be obtained by any number of approaches. Minimally, an existing discrete variable may be chosen as a segmentation, also called "a priori" segmentation. At the other extreme, a research project may be commissioned to collect data on many attributes and use statistical analyses to derive a segmentation, also called "post-hoc" segmentation. In between, qualitative knowledge of the market based on experience may be used to identify divisions that are likely to be useful.

Common statistical techniques for segmentation analysis include:

• Clustering algorithms such as K-means or other Cluster analysis
• Statistical mixture models such as Latent Class Analysis
• Ensemble approaches such as Random Forests

Latent class analysis and k-means analysis may be viewed as identifying new variables that maximize the sum of mutual information between the segmentation variable and a set of basis variables.