What: K2S builds models that create groups or clusters of objects (for example customers) that have a similar profile.

Why: Clustering provides the description of a group with similar behaviour to implement incentives causing a similar reaction from the group members. By defining these groups, organisations are enabled to efficiently address the needs of their target groups.

How: K2S builds models implementing a mapping between a set of descriptive attributes (model inputs) and the cluster ID (model output). The distance used to determine the closest centre is generated by K2C. The segmentation can either be supervised or unsupervised depending on availability of a business question. The output of a clustering model is presented via basic statistics on clusters:

• Frequency: percentage of population gathered in the cluster.

• % of “label” in classification case for a binary target: label percentage in the cluster, where label is the binary target least frequent category.

• Target Mean in regression case for continuous target: target mean value for data assigned to the cluster.

• Cross statistics per cluster will compare per variable the cluster value and the overall population value.

Benefits for the business user: the business user can easily create meaningful market segmentations. Each cluster is homogeneous with respect to the entire set of variables. Clusters will display characteristics of those with high response rate and poor response rate.

Two people performing this segmentation using KXEN methods will obtain the same results.

Benefits for the Data Mining expert: K2S offers automatic and robust segmentation with respect to a given business question.

Benefits for the Integration specialist and IT: K2S allows building clusters on large data sets within minutes.

Example:

Hot line management usage scenario.

• Collect information about customers, and associate them with the average number of phone calls to the hot line

• Train a K2S model that generates definitions of groups (clusters) of customers with roughly same behaviour towards the hot line

• Associate each customer with its group “colour” and customise hot-line scenario per group.