Mining Your Own Business in Telecoms Using DB2 Intelligent Miner for Data

Chapter 4. How to discover the characteristics of your customers 61

If we give the demographic clustering algorithm a similarity threshold and do not

limit the number of clusters that it can produce, it will keep trying to find the

minimum number of clusters that satisfy the similarity threshold, because this

also maximizes the Condorcet value.

4.6 Interpreting the results

In the previous section we looked at the steps we have to follow to get our mining

results using two different clustering techniques. The

sixth stage in our generic

mining method

is to interpret the results that we have obtained and determine

how we can map them onto our business. When you are first confronted with the

cluster results the first question that you are going to ask is “What does it all

mean?”. In this section we describe how to understand and read the visualized

segmentation output and how to interpret the results from the neural clustering

technique that has generated nine different segments of customers.

4.6.1 How to read and evaluate the results

The cluster techniques that we have used both produce results that can be

displayed graphically. Although the visualized results are important in giving us

an overall impression of what is happening, evaluating the result with the

business point of view should not be neglected.

How to read the result?

The graphic output of neural clustering is shown in Figure 4-2. Each segment (or

cluster) is interpreted with its own behaviors and demographic and additional

variables.

Note: Condorcet like similarity has value in the ranges zero to one and is a

measure of how similar a customer is to other customers within the cluster and

how dissimilar they are to customers in other clusters. It has a value of 1 when

all customers in a cluster are identical and where there are no customers

outside the cluster that have the same characteristic. A value of zero indicates

that the customers are distributed randomly among the clusters. The

condorcet value can be calculated for all the customer variables or for each

variable separately.

62 Mining Your Own Business in Telecoms Using DB2 Intelligent Miner for Data

The small number in the right most of each strip indicates the segment identifier

(ID). Derived segments are displayed in each strip ordered by the relative size

which is indicated by the number in left most vertical. In each strip, variables are

presented either as a histogram if it is numeric, or a pie chart if it is categorical,

and the order of variables that appear in each strip gives an idea to the user of

which variables are more significantly used when the segment is defined.

The basic idea of interpreting the visualized variables in the strip is to examine

the variables and which category or value range presents the largest difference

from the population average.

Figure 4-2 Customer segmentation output using neural clustering

Chapter 4. How to discover the characteristics of your customers 63

The pie chart shows the distribution of categorical variables for customers within

the segment (inner circle) which can be compared with the distribution of the all

customers (outer circle). In a similar fashion the histogram shows the distribution

of numeric variables for customers within the particular cluster as red

(non-shaded) which can be compared with the distribution for all customers

which is gray (shaded).

For example as shown in Figure 4-3, the cluster has 40:60 ratio (inner circle) of

female and male which can be compared with the whole population (45:55 —

outer annulus).

Figure 4-3 Pie chart visualization - Gender

This visualization method is very intuitive and helps business users to

understand the whole segmentation, fast and easy.

How to evaluate the result?

The most important criteria to evaluate a derived segmentation should be that

the results always make sense from a business point of view. A good way of

checking this is to go through the different segments and try to give them a

descriptive name which is meaningful in a business perspective.