A
Binning

This appendix provides a detailed example of binning.

Table A-1 displays original data before binning. Table A-2 shows the bin boundaries for numeric data; Table A-3 shows bin boundaries for categorical data. Table A-4 shows the results of binning.

Table A-1 Binning Illustration: Data before Binning

PERSON_ID	AGE	WORK CLASS	EDUCATION	MARITAL_STATUS	OCCUPATION
2	27	Private	HS-grad	Married	Crafts
8	46	Private	Bach.	Separ.	Prof.
10	34	Private	HS-grad	Separ.	Agricultural
11	23	Sta-gov	< Bach.	NeverM	Cleric.
41	30	Private	< Bach.	Married	Sales

Table A-2 Binning Illustration: Bin Boundaries for Numeric Data

COLUMN_NAME	LOWER_ BOUNDARY	UPPER_BOUNDARY	BIN_ID	DISPLAY_NAME
AGE	17	24.3	1	17-24.3
AGE	24.3	31.6	2	24.3-31.6
AGE	31.6	38.9	3	31.6-38.9
AGE	38.9	46.2	4	38.9-46.2
AGE	46.2	53.5	5	46.2-53.5

Table A-3 Binning Illustration: Bin Boundaries for Categorical Data

COLUMN_NAME	CATEGORY	BIN_ID	DISPLAY_NAME
WORKCLASS	Loc-gov	1	Government
WORKCLASS	Fed-gov	1	Government
WORKCLASS	Sta-gov	1	Government
WORKCLASS	Private	2	Others
EDUCATION	HS-grad	1	HS-grad
EDUCATION	< Bach.	2	< Bach.
EDUCATION	Bach.	3	Bach.
EDUCATION	Masters	4	Masters
MARITAL_STATUS	Married	1	Married
MARITAL_STATUS	NeverM	2	NeverM
MARITAL_STATUS	Divorc.	3	Divorc.
MARITAL_STATUS	Widowed	4	Widowed
MARITAL_STATUS	Separ.	5	Separ.
OCCUPATION	Prof	1	Prof
OCCUPATION	Crafts	2	Crafts
OCCUPATION	Exec.	3	Exec.
OCCUPATION	Sales	4	Sales
OCCUPATION	Cleric	5	Cleric
OCCUPATION		6	Other_occ

Table A-4 Binning Illustration: Assignment of Original Data to Bins

PERSON_ID	AGE	WORK CLASS	WEIGHT	EDUCATION	MARITAL_STATUS	OCCUPATION
2	2	2	2	1	1	2
8	4	2	1	3	5	1
10	3	2	1	1	5	6
11	1	1	1	2	2	5
41	2	2	2	2	1	4

A.1 Use of Automated Binning

The Java interface supports automated binning. An important advantage of automated binning is that it allows ODM to handle raw data. Automated binning also allows initial exploration of problems about which there is little or no information to guide binning decisions.

Currently automatic binning requires closed intervals for numerical bins. This can result in certain values being ignored. For example, if the salary range in the build data table is 0 to 1,000,000, any salary greater than 1,000,000 is ignored when the model is applied. If you are trying to identify likely purchasers of a high-end consumer product, attributes indicating the wealthiest individuals are likely to be deleted, and you probably won't find the best targets. Manual binning has the option of making extreme bins open-ended, that is, with infinite boundaries.