Cluster Analysis vs. Market Segmentation
Introduce cluster analysis and market segmentation by discussing:
* Concept of cluster analysis and basic ideas and algorithms
* Concept of market segmentation and basic ideas
* Comparison of these two approaches
Cluster Analysis Algorithms
"There appear to be more algorithms for clustering data than data to analyze."
What is Cluster Analysis?
- A cluster is a group of similar objects (cases, points, observations, examples, members, customers, patients, locations, etc)
- Cluster Analysis is a set of data-driven partitioning techniques designed to group a collection of objects into clusters, such that
- the number of groups (clusters) as well as their forms are unknown
- the degree of association or similarity
- is strong between members of the same cluster
- is weak between members of different clusters
- The nature of Cluster Analysis is data exploration that is conducted in repetitive fashion. Clusterization is not a single grouping, but the process of getting well-interpretable groups of objects under consideration.
What is not Cluster Analysis?
- Supervised classification, for example, Discriminant Analysis, Naive Bayes Classifier, Support Vector Machines, etc.
- Have class label information
- Simple segmentation
- Doctors segmentation by specialty, assuming that each doctor's specialty is known
- Customer segmentation by sex, education level, geography and response rate (assuming that these customer attributes are known)
- Results of a query (groupings are the outcome of an external specification)
Supervised vs. Unsupervised
Cluster analysis is the product of at least two different quantitative fields: statistics and machine learning
- Unsupervised is learning from raw data (no examples of correct classification). In other words, class label information is unavailable.
- No measure of success
- Heuristic arguments for judgments
- Lots of methods developed
- Supervised is learning from data where the correct classification of examples is given (class label information is available)
- Market segmentation is one of the most fundamental strategic marketing concepts:
- grouping people (with the willingness, purchasing power, and the authority to buy) according to their similarity in several dimensions related to a product under consideration.
- The better the segments chosen for targeting by a particular organization, the more successful the organization is in the marketplace. The objectives are accurately predict the needs of customers and improve the profitability.
Variables used in market segmentation
- Home ownership, etc.
- Buying motives, etc.
- Brand Loyalty
- City size
- Rural vs. Urban, etc.
Market Segmentation and Cluster Analysis
- Help marketers discover distinct groups in their customer bases, and then use this knowledge to develop targeted marketing programs
- The underlying definition of cluster analysis procedures mimic the goals of market segmentation:
- to identify groups of respondents that minimizes differences among members of the same group
- highly internally homogeneous groups
- while maximizing differences between different groups
- highly externally heterogeneous groups
- Market Segmentation solution depends on
- variables used to segment the market
- method used to arrive at a certain segmentation
Criteria for Successful Market Segmentation
- Can we see clear differences between segments?
- Are the segments large enough to warrant separate marketing targeting?
- Can we reach our customers?
- Do our segments stable over a certain period of time?
- Is the response to our marketing effort segment specific?
- Does the segmentation provide direction for marketing efforts?
Types of Clustering
- Partitional clustering
- A division of objects into non-overlapping subsets (clusters) such that each object is in exactly one cluster
- Hierarchical clustering
- A set of nested clusters organized as a hierarchical tree
Other Distinctions Between Different Forms Of Clustering
- Different treatment of object characteristics vs. even treatment
- Characteristics are subdivided into two groups: dependent variable and independent variables (Classification and Regression trees)
- There is no such subdivision (K-means)
- Model-based vs. Non-model-based
- A model is hypothesized for each of the clusters and the idea is to find the best fit of that model to each cluster (Latent Class Clustering)
Limitations and Problems of Traditional Cluster Analysis Methods
- Need to specify K (number of clusters) in advance
- Applicable only for interval variables (only numeric data)
- Has problems when clusters are of differing
- Non-globular shapes
- Unable to handle noisy data and outliers
Latent Class Cluster Analysis (LCCA)
- LCCA is a model-based approach:
- The statistical model is postulated for the population from which the data sample is obtained
- LC models do not rely on the traditional modeling assumptions (linearity, normality, homogeneity)
- It is assumed that a mixture of underlying probability distributions generates the data
- LC models include a K-category latent variable, each category represents a cluster
- Objects are classified into clusters based upon membership probabilities that are estimated directly from the data
Advantages of Latent Class Cluster Analysis (LCCA)
- The optimal number of clusters is determined as a result of LCCA, using rigorous statistical tests
- No decisions have to be made about the scaling of the observed variables
- Variables may be continuous, nominal, ordinal, count, or any combination of these
Theory and Cluster Analysis
- Is clustering a theory?
- A theory could be true or false
Unlike a theory, clustering is neither true nor false, and should be judged largely on the interpretability and usefulness of results
- No measure of success
Heuristic arguments for judgments
- Selection of right method is a problem
However, a clustering may be useful for suggesting a theory, which could then be tested
- Leonard Kaufman and Peter Rousseeuw (2005), Finding Groups in Data: An Introduction to Cluster Analysis, Wiley Series in Probability and Statistics, 337 p.
- Mark Aldenderfer and Roger Blashfield (1984), Cluster Analysis (Quantitative Applications in the Social Sciences), SAGE Publications, Inc., 90 p.
- Brian Everitt, Sabine Landau and Morven Leese (2001) Cluster Analysis, Oxford University Press, 248 p.
- Marketing Segmentation (http://www.beckmanmarketing8e.nelson.com/ppt/chapter03.pps
Application of clustering and customer segmentation to survey data
Case study: background, objectives, and methodology
- A producer and distributor of health and beauty products has launched a new product. The product can only be orderedon the website.
- In six months an internet survey was conducted. Only three simple questions were asked:
- How many adults are in your household?
- How many of them adopted the product?
- How many of them did not adopt the product?
- When the total number of adopters and non-adopters is less than the number of adults in a household, the difference is treated as the number of unknowns. There are some other situations when the number of unknowns makes sense to introduce.
- The client asked us to analyze the survey data (obviously it is not the most informative survey BI Solutions has dealt with).
- The objectives of the study was to extract as much useful information as possible from the survey data in order to understand the distribution and the usage of the product among households, associate with each household a corresponding likelihood of adoption, and develop a methodology to employ this info in marketing programs.
- Methodology: synergy of cluster analysis of proportional data and intuitive segmentation.
Clustering of households
- We calculated the following three variables:
- P1 is the proportion of customers in a household with unknown product adaption behavior
- P2 is the proportion of customers in a household who adopted the product
- P3 is the proportion of customers in a household who did not adopt the product
- Therefore, each household is characterized by a point in a three-dimensional proportion space. Once again, it was the only available information (that we got from the client).
- We decided to employ a synergy of cluster analysis and customer segmentation. Six clusters were identified as the result of K-means clustering.
- Variables are important in K-means clustering.
Household representation in three dimensional proportion space
Data Mining and Related Fields
Clusters of households: mean value of proportions and cluster interpretation
Cluster Profiling: good separation and good interpretability
Households as a set of three proportions/percentages
Cluster /Sub-cluster profile
Likelihood of new product adoption
- Customer profiling
- Data enrichment
- Data enrichment (ZIP level census data)
- Usage other health/beauty products (household level data)
- Estimation of the likelihood of the product adoption by data mining predictive analysts / scoring households with unknown purchasing behaviour
- Identifying customers with high likelihood of the product adoption for targeting
- Developing program for increasing up-sell and cross-sell
- Developing program for customer retention
- Spatial clustering of potential and real customers