TRAINING COURSE : REGRESSION

* How much will this family spend on their vacations ?

* How long will this driver keep his car ?

* What is a "fair" salary for this position ?

* What sales level can be expected for this new store ?

These questions, and many others, call for predicting the value of a number in a given circumstance. Answering this type of question is doing prediction, or regression.
 

Data Mining has many prediction techniques. They differ widely in terms of performances and operational characteristics. This 1 or 2 day training course (see outline below) reviews the most popular prediction available in most Data Mining software.

Outline of the course

The general problem of prediction

The "best fitting" function

Why it is difficult to find the best fitting function

Data dispersion

Complexity of the function

Uncertainties about the predictions, generalization capacity

Linear Regression                                

Least Squares line

Overall model quality, Rē

Assumptions and limitations

Linearity

Normality and independence of the errors

Homoscedasticity

Multiple Linear Regression   (MLR)

What's "on top" of (simple) Linear Regression ?

Chosing the predictors

Variables redundency and model stability

Multiple and partial correlation

Chosing the predictors

The" curse of dimensionality"

Stepwise methods : ascending, descending, mixed.

Hunting down variables redundency

Neural Networks (Supervised)

What do Neural Networks actually do ?

The main types of Neural Networks

The Multilayer Perceptron

RBF networks

Pros and cons of Neural Networks

High prediction accuracy

Tests and interpretability of the parameters