TRAINING COURSE "CLUSTERING AND SEGMENTATION in Data Mining"

Are your customers' profiles just about uniformly distributed in the "space" defined by their attributes ? Or do they rather tend to "cluster" in a small number of compact and homogenous groups ?

Do their buying patterns can also be described in terms of "segments", each segment being well defined by a set of values of their attributes ?
 

Many companies do not know. Yet, this kind of information is very useful. It allows identifying a few specific customer profiles that can then be targeted by a well adapted, and therefore more efficient promotional effort.

These questions, and many others, belong to the two similar, but different problems : clustering and segmention. These activities are among the most popular in Data Mining, as they directly provide usable information where there was only data.

This one day training course (see outline below) reviews the main clustering and segmentation techniques that are in most major Data Mining software.

Outline of the course

General principles of clustering

The two-dimension example

Similarity and distance between individuals

How to choose a "clustering quality" criterion

How many classes?

Hierarchical clustering                

The agregation principle

Dendogram

The decisions you have to make

Distance between individuals

Distance between classes

Agregation criterion

Pros and cons of hierarchical clustering

Chosing the number of clusters

Influence of the parameters

Robustness

How to assign new individuals to clusters

Computation time

K-means

Principle of k-means

Prototypes and barycentres

Why are there several solutions ?

Pros and cons of k-means

Chosing the number of clusters

How to assign new individuals to clusters

Why are there several solutions ?

Problems with initialization

Kohonen maps

Basic principles of Kohonen Maps

Architecture

Kohonen training algorithm

Kohonen Maps and PCA

Kohonen Maps and clustering

"Neurons" and micro-classes

From micro- to macro-classes

How to assess the success of the training phase

Pros and cons of Kohonen Maps

Non linearity

"Topology" and proximity between clusters

Parameters adjustment

Assessing the quality of the map

Fuzzy Clustering

Why "Fuzzy" clustering ?

An individual belongs to several clusters

Fuzzy clustering techniques

How to interpet fuzzy clusters

Segmentation Trees

Segmentation and clustering

Principles of Segmentation Trees

"Discriminating Power" of an attribute

Recursive partitioning of a data base

Stopping the Tree growth

The various splitting criteria

CHAID, entropy,  Gini index

Categorical and numerical variables