Quartile Binning

This is an excerpt from Dr. Ham's premier book "Oracle Data Mining: Mining Gold from your Warehouse".

We see that ODMr recognized that the two numerical data attributes AGE and YRS_RESIDENCE should be binned, and discretized the data so that these fields were categorized into 3 different bins: 1, 2, and 3.

When we click on the Options button in the Discretize section, we will find that our options are Quantile Binning, Equal Width Binning, and None. 

We can illustrate the difference between the quantile and equal width binning by using the discretize wizard.  Below is the histogram for the attribute AGE in the MINING_DATA_BUILD_V_US case dataset using the equal width binning strategy. 

Each group in the histogram view is composed of age values in increasing increments of 7.3 years.  As age increases, the number of customers in the bins decreases, from a maximum of 19.49% in group 3 to 0.22% in group 9.  This type of distribution that “tails off” is not a good choice for data mining analysis.  You want a more uniform distribution of ages across all groups, as in the quantile binning shown below. 

Using the Discretize Transform Wizard

You can view the histograms of data using different binning methods in the discretize wizard.  We will illustrate the use of the discretize wizard by customizing the age ranges in a new binning strategy. 

In Step 3 of the Wizard, note that there are 3 attributes identified as numerical mining types:  AGE, CUST_ID, and YRS_RESIDENCE.  In this step you can change the mining type if necessary by clicking “numerical” or “categorical” and choosing the appropriate type. 

Numerical data can be binned.  Click next to continue the wizard. 

For example if you pick Adaptive Bayes Networkfor AGE, the binning wizard bins the data into 5 groups:  < 26, 26-34, 34-42, 42-51, and >51.  The Naïve Bayes defaults to 3 groups:  <32, 32-44, and >44.